System and method for assessing wellness

Abstract

To provide a system, method, and apparatus for assessing pet wellness.SOLUTION: This method includes the steps of: receiving data related to a pet; based on the data, determining one or more health indicators of the pet; and, based on the one or more health indicators, performing a wellness assessment of the pet. Also, the method further includes the steps of: determining a recommendation to a pet owner based on the wellness assessment; and transmitting the recommendation to a mobile device of the pet owner, where the recommendation is displayed at the mobile device to the pet owner.SELECTED DRAWING: None

Description

Related Applications

[0001] This application claims the benefit of priority to U.S. Patent Application No. 62 / 867,226, filed June 26, 2019, U.S. Patent Application No. 62 / 970,575, filed February 5, 2020, and U.S. Patent Application No. 63 / 007,896, filed April 9, 2020, all of which are incorporated herein by reference in their entireties. [Technical field]

[0002] Embodiments described in this disclosure relate to data analysis. For example, some non-limiting embodiments relate to data analysis of pet activity or other data. [Background technology]

[0003] Portable and / or wearable devices have been equipped with various hardware and software components that may assist in tracking a person's location. For example, portable devices may communicate with a Global Positioning System (GPS) to facilitate determining their location. More recently, portable and / or wearable devices may include sensors that go beyond simple location tracking and assist in monitoring human activity. Data obtained from tracked location and / or monitored activity may be collected, analyzed, and displayed. For example, portable and / or wearable devices may be used to track the number of steps a person takes in a pre-set period of time. The number of steps may be displayed in a user graphic interface of the portable or wearable device. Summary of the Invention [Problem to be solved by the invention]

[0004] With the increasing emphasis on pet safety and health, the need to monitor pet behavior has increased. Thus, there is a continuing demand in the pet product industry for systems and / or methods for tracking and monitoring pet activity. In particular, there is a demand for wearable pet devices that can accurately track the location of a pet while monitoring the pet's activity. [Means for solving the problem]

[0005] To address the above-mentioned problems, the present disclosure presents systems, methods, and devices that can be used to analyze data. For example, some non-limiting embodiments can be used to monitor and track the activities of pets.

[0006] In a non-limiting embodiment, the present disclosure describes a method for monitoring pet activity. The method includes monitoring a location of a wearable device. The method also includes determining that the wearable device has left a geofenced zone based on the location of the wearable device. The method further includes instructing the wearable device to turn on an indicator after determining that the wearable device has left the geofenced zone. The indicator can be at least one of a lighting device, an audio device, or a vibration device. The method can also include determining that the wearable device has entered the geofenced zone, and turning off the indicator after the wearable device has entered the geofenced zone.

[0007] In a non-limiting embodiment, the present disclosure describes a method for monitoring a pet's activity. The method includes receiving data related to a pet from a wearable device comprising a sensor. The method also includes determining one or more health indicators of the pet based on the data, and performing a health assessment of the pet based on the one or more health indicators of the pet. The method may also include transmitting the health assessment of the pet to a mobile device. The health assessment of the pet may be displayed to a user on the mobile device. The method may be performed by the wearable device, one or more servers, a cloud computing platform, and / or any combination thereof.

[0008] In some non-limiting embodiments, the disclosure describes a method that may include receiving data at a device. The method may also include analyzing the data using two or more layer modules, each layer module including at least one of a many-to-many approach, striding, downsampling, pooling, multi-scaling, or batch normalization. The method may also include determining an output based on the analyzed data. The data may include at least one of financial data, cybersecurity data, electronic health records, health data, image data, video data, audio data, human activity data, pet activity data, and / or any combination thereof. The output may include one or more of a health assessment, health recommendations, financial forecasts, security recommendations, image or video recognition, voice recognition, and / or any combination thereof. The determined output may be displayed on a mobile device.

[0009] In a non-limiting embodiment, the present disclosure describes a method for assessing the health of a pet. The method may include receiving data related to a pet, determining one or more health indicators of the pet based on the data, and performing a health assessment of the pet based on the one or more health indicators. The method may also include providing or determining a recommendation to a pet owner based on the health assessment, and transmitting the recommendation to a mobile device of the pet owner, where the recommendation is displayed on the mobile device of the pet owner.

[0010] In a non-limiting embodiment, a device for monitoring a pet's activity may include at least one memory comprising computer program code and at least one processor. The computer program code, when executed by the at least one processor, may be configured to cause the device to receive data regarding a pet from a wearable device comprising a sensor. The computer program code, when executed by the at least one processor, may also be configured to cause the device to determine one or more health indicators of the pet based on the data, and to perform a health assessment of the pet based on the one or more health indicators of the pet. The computer program code, when executed by the at least one processor, may also be configured to cause the device to transmit a health assessment of the pet to a mobile device. The health assessment of the pet is displayed to a user on the mobile device.

[0011] A non-limiting embodiment may be directed to a wearable device. The wearable device may include a housing having a top cover. The housing may also include a base coupled to the top cover. The housing may also include a sensor for monitoring data related to a pet and a transceiver for transmitting the data related to the pet. The housing may also include an indicator, the indicator being at least one of a lighting device, a sound device, a vibration device, and / or any combination thereof.

[0012] According to non-limiting embodiments, at least one non-transitory computer readable medium is provided that stores instructions that, when executed by hardware, perform a process according to the methods disclosed herein. In some non-limiting embodiments, the apparatus may include a computer program product encoded with instructions for processing data of tested pet products according to the methods. In other embodiments, the computer program product may be encoded with instructions for performing a process according to the methods disclosed herein.

[0013] It is to be understood that both the foregoing general description and the following detailed description are exemplary and are intended to provide further explanation of the disclosed subject matter claimed. [Brief description of the drawings]

[0014] The above and other objects, features, and advantages of the present disclosure will become apparent from the following description of the embodiments, as illustrated in the accompanying drawings, in which like reference characters refer to like parts throughout, and in which the drawings are not necessarily to scale, with emphasis instead being placed on illustrating the principles of the present disclosure. [Figure 1] 1 illustrates a system used to track and monitor pets, according to a non-limiting embodiment. [Diagram 2] 1 illustrates a device that may be used to track and monitor pets, according to a non-limiting embodiment. [Diagram 3] FIG. 1 is a logical block diagram illustrating a device that may be used to track and monitor pets according to a non-limiting embodiment. [Figure 4] FIG. 1 is a flow diagram illustrating a method for tracking a pet, according to a non-limiting embodiment. [Diagram 5] FIG. 1 is a flow diagram illustrating a method for tracking and monitoring a pet, according to a non-limiting embodiment. [Figure 6] FIG. 6 illustrates two example deep learning models according to non-limiting embodiments. [Figure 7a] 1 illustrates a model architecture according to a non-limiting embodiment. [Figure 7b] 1 illustrates a model architecture according to a non-limiting embodiment. [Figure 7c] 1 illustrates a model architecture according to a non-limiting embodiment. [Figure 8] 1 illustrates an example model according to a non-limiting embodiment. [Figure 9] 9 illustrates an embodiment of the model shown in FIG. 8. [Figure 10] 9 illustrates an example architecture that may consist of one or more of the models shown in FIG. 8. [Figure 11] 4 illustrates example model parameters according to a non-limiting embodiment. [Figure 12] 1 illustrates an example of a representative model training run according to a non-limiting embodiment. [Figure 13] 1 illustrates the performance of an example model according to a non-limiting embodiment. [Figure 14] 1 illustrates a heatmap showing model performance according to a non-limiting embodiment. [Figure 15] 1 illustrates performance metrics of a model according to a non-limiting embodiment. [Figure 16] 1 shows the performance of an n-fold ensemble ms-C / L model according to a non-limiting embodiment. [Figure 17] 13 illustrates the effect of varying the sliding window length used in the interference step according to a non-limiting embodiment. [Figure 18] 14 illustrates the performance of one or more models according to non-limiting embodiments based on multiple sensors. [Figure 19] 1 illustrates a performance analysis of a model according to a non-limiting embodiment. [Figure 20] FIG. 1 illustrates a flow diagram of a process for assessing a pet's health, according to a non-limiting embodiment. [Figure 21] 1 illustrates example steps performed during a process for assessing a pet's health according to a non-limiting embodiment. [Figure 22-1] 1 illustrates example steps performed during a process for assessing a pet's health according to a non-limiting embodiment. [Figure 22-2] A continuation of Figure 22-1 is shown. [Figure 23] 1 illustrates example steps performed during a process for assessing a pet's health according to a non-limiting embodiment. [Figure 24] 1 illustrates example steps performed during a process for assessing a pet's health according to a non-limiting embodiment. [Figure 25A] FIG. 1 shows a flow diagram of a process for assessing a pet's health according to a non-limiting embodiment. [Figure 25B] FIG. 1 shows a flow diagram of a process for assessing a pet's health according to a non-limiting embodiment. [Figure 26] FIG. 1 shows a flow diagram of a process for assessing a pet's health according to a non-limiting embodiment. [Figure 27] FIG. 1 shows a perspective view of a collar with a tracking device and a band according to an embodiment of the disclosed subject matter. [Figure 28] 28 shows a perspective view of the tracking device of FIG. 27 in accordance with the disclosed subject matter. [Figure 29] FIG. 28 illustrates a front view of the tracking device of FIG. 27 in accordance with the disclosed subject matter. [Diagram 30] 28 shows an exploded view of the tracking device of FIG. 27. [Diagram 31] 28 shows a left side view of the tracking device of FIG. 27, with the right side view being identical to the left side view. [Diagram 32] 28 shows a top view of the tracking device of FIG. 27, with the bottom view being identical to the top view. [Diagram 33] FIG. 28 shows a rear view of the tracking device of FIG. [Diagram 34] FIG. 1 shows a perspective view of a tracking device according to another embodiment of the disclosed subject matter. [Diagram 35] FIG. 35 illustrates a front view of the tracking device of FIG. 34 in accordance with the disclosed subject matter. [Diagram 36] 35 shows an exploded view of the tracking device of FIG. 34. [Figure 37] FIG. 35 illustrates a front view of the tracking device of FIG. 34 in accordance with the disclosed subject matter. [Figure 38] 35 shows a left side view of the tracking device of FIG. 34, with the right side view being identical to the left side view. [Figure 39] 35 shows a top view of the tracking device of FIG. 34, with the bottom view being identical to the top view. [Diagram 40] 35 shows a rear view of the tracking device of FIG. 34. [Diagram 41] 1 illustrates a rear view of a cable-connectable tracking device in accordance with the disclosed subject matter. [Diagram 42] 1 depicts a collar having a receiving plate for receiving a tracking device in accordance with the disclosed subject matter. [Diagram 43] 1 depicts a pet wearing a collar in accordance with an embodiment of the disclosed subject matter. [Diagram 44] 1 depicts a pet wearing a collar in accordance with an embodiment of the disclosed subject matter. [Diagram 45] 1 depicts a collar receiving plate and / or support frame for receiving a tracking device according to another aspect of the disclosed subject matter. [Figure 46] 1 depicts a collar receiving plate and / or support frame for receiving a tracking device according to another aspect of the disclosed subject matter. [Figure 47] 1 depicts a collar receiving plate and / or support frame for receiving a tracking device according to another aspect of the disclosed subject matter. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] There is a need for systems, methods, and devices that can monitor and track the activities of pets. The subject matter of the present disclosure addresses this need, as well as other needs related to pet health. Specifically, data regarding the tracking or monitored activities of a pet may be collected and used to detect any possible health risks associated with the pet. A summary of the identified possible health risks and the collected data may be sent to and / or displayed to or by the pet owner.

[0016] U.S. Patent Application No. 15 / 291,882, now U.S. Patent No. 10,142,773, U.S. Patent Application No. 15 / 287,544, U.S. Patent Application No. 14 / 231,615, U.S. Provisional Patent Applications Nos. 62 / 867,226, 62 / 768,414, 62 / 970,575, and 63 / 007,896, and U.S. Design Applications Nos. 29 / 696,311 and 29 / 696,315 are incorporated herein by reference. All subject matter disclosed in the above-referenced applications, including the specifications, claims, and figures, is incorporated herein by reference.

[0017] The present disclosure is described more fully below with reference to the accompanying drawings, which form a part hereof and which show, by way of example, embodiments. However, the subject matter can be embodied in many different forms, and the claimed subject matter is not intended to be construed as being limited to any embodiments set forth herein. The embodiments are provided merely as examples. A reasonably broad scope of the claimed subject matter is also intended. Among other things, the subject matter can be embodied as a method, apparatus, component, or system. Thus, the embodiments can take the form of, for example, hardware, software, firmware, or any combination thereof (other than pure software). Thus, the following detailed description is not intended to be construed in a limiting sense.

[0018] In the detailed description herein, references to "an embodiment," "one non-limiting embodiment," "in various embodiments," and the like indicate that the described embodiment may include a particular feature, structure, or characteristic, but all embodiments may not necessarily include that particular feature, structure, or characteristic. Moreover, such phrases do not necessarily refer to the same embodiment. Moreover, one of ordinary skill in the art is said to know that when a particular feature, structure, or characteristic is described in connection with one embodiment, it also affects such feature, structure, or characteristic in connection with other embodiments, whether or not expressly stated. After reading the description, it will be apparent to one of ordinary skill in the art how to implement the present disclosure in other embodiments.

[0019] Generally, terms may be understood, at least in part, from their usage in context. For example, terms such as "and," "or," and "and / or" as used herein may include various meanings that may depend, at least in part, on the context in which they are used. Typically, "or," when used in a list such as A, B, or C, is intended to mean A, B, and C in an inclusive sense, and A, B, or C in an exclusive sense. Also, as used herein, the term "one or more" may be used to describe any feature, structure, or characteristic in a singular sense, or may be used to describe a combination of features, structures, or characteristics in a plural sense, depending, at least in part, on the context. Similarly, terms such as "a," "an," or "the" may be understood to convey singular or plural usage, depending, at least in part, on the context. Also, it is understood that the term "based on" is not necessarily intended to convey an exclusive set of factors, and may allow for the presence of additional factors not necessarily specified, depending, at least in part, on the context.

[0020] As used herein, the terms "comprises," "comprising," or variations thereof are intended to mean a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises recited elements not only includes those elements, but may also include other elements that are not specific to the process, method, article, or apparatus and are not recited.

[0021] The present disclosure is described below with reference to block diagrams and operational diagrams of methods and apparatus. It will be understood that each block of the block diagrams or operational diagrams, and combinations of blocks in the block diagrams or operational diagrams, can be implemented by analog or digital hardware and a number of computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, an ASIC, or other programmable data processing apparatus that changes the functions as detailed herein, and the instructions executed by the processor of the computer or other programmable data processing apparatus perform the functions / operations noted in the block diagrams or operational blocks. In some other embodiments, the functions / operations noted in the blocks can be performed out of the order noted in the operational diagrams. For example, two consecutive blocks can be executed substantially simultaneously, or the blocks can sometimes be executed in a reverse order depending on the functions / operations involved.

[0022] These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, ASIC, or other programmable data processing apparatus that modifies its functionality for a specific application, and the instructions executed by the processor of the computer or other programmable data processing apparatus perform the functions / operations noted in the block diagrams or operational blocks, thereby modifying those functions depending on the embodiment.

[0023] For purposes of this disclosure, a computer-readable medium (or computer-readable storage medium) stores computer data, which may include computer program code (computer-executable instructions) in machine-readable form that is executable by a computer. By way of example, and not limitation, a computer-readable medium may consist of a computer-readable storage medium for the tangible or fixed storage of data, or a communication medium for the transitory interpretation of signals containing code. As used herein, a computer-readable storage medium refers to a physical or tangible storage (as opposed to a signal), including, but not limited to, volatile and non-volatile removable and non-removable media implemented in any method or technology for the tangible storage of information, such as computer-readable instructions, data structures, program modules, and the like. A computer readable storage medium includes, but is not limited to, RAM, ROM, EPROM, EEPROM, flash memory or other solid-state memory technology, CD-ROM, DVD or other optical storage devices, magnetic cassettes, magnetic tapes, magnetic disk storage devices or other magnetic storage devices, or any other physical or tangible medium that can be accessed by a computer or processor and used to tangibly store desired information or data or instructions.

[0024] For purposes of this disclosure, the term "server" should be understood to refer to a service point that provides processing, database, and communication facilities. By way of example, and not by way of limitation, the term "server" may refer to a single physical processor with associated communication and data storage and database capabilities, or may refer to a networked or clustered collection of processors, such as a flexible computer cluster, associated network and storage devices, and operating software and one or more database systems and application software that support the services that the server provides. A server may be, for example, a cloud-based server, a cloud computing platform, or a virtual machine. Servers may vary widely in configuration or capabilities, but generally a server may include one or more central processing units and memory. A server may also include one or more mass storage devices, one or more power sources, one or more wired or wireless network interfaces, one or more input / output interfaces, or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, and the like.

[0025] For purposes of this disclosure, a "network" should be understood to refer to a network that may couple devices such that communications may be exchanged, for example, between a server and a client device or between other types of devices (including between wireless devices coupled via a wireless network). A network may also include mass storage devices, such as, for example, a network attached storage (NAS), a storage area network (SAN), or other forms of computer-readable or machine-readable media. A network may include the Internet, one or more local area networks (LANs), one or more wide area networks (WANs), wired connections, wireless connections, cellular, or any combination thereof. Similarly, sub-networks that may use different architectures or conform to or be compatible with different protocols may co-operate within a larger network. Various types of devices may be made available, for example, to provide interoperability of different architectures or protocols. As one example, a router may provide a link between separate and independent LANs.

[0026] The communications links or channels may include, for example, analog telephone lines such as twisted pair wires, coaxial cables, fully or partially digital lines including T1, T2, T3, or T4 type lines, Integrated Services Digital Networks (ISDN), Digital Subscriber Lines (DSL), wireless links including satellite links, or other communications links or channels that may be known to those skilled in the art. Also, computer devices or other related electronic devices may be remotely connected to the network, for example, via wired or wireless lines or links.

[0027] For purposes of this disclosure, a "wireless network" should be understood to combine a client device with a network. The wireless network may use a standalone network, a mesh network, a wireless local area network (WLAN), a cellular network, etc. The wireless network may further include a system of terminals, gateways, routers, etc., coupled by wireless links, etc., that may move freely and randomly or organize themselves arbitrarily, and the network configuration may change, sometimes even rapidly.

[0028] A wireless network may use multiple network access technologies including Wi-Fi, Long Term Evolution (LTE), WLAN, Wireless Router (WR) mesh, or second, third, fourth, or fifth generation (2G, 3G, 4G, or 5G) cellular technologies, etc. A network access technology may enable wide area coverage for devices, such as client devices, with different mobility.

[0029] For example, a network may enable RF or wireless communications via one or more network access technologies such as Global System for Mobile Communications (GSM), Universal Mobile Telecommunications System (UMTS), General Packet Radio Service (GPRS), Evolved Data GSM Environment (EDGE), 3GPP LTE, LTE Advanced, Wideband Code Division Multiple Access (WCDMA), Bluetooth, 802.11b / g / n, etc. A wireless network may include virtually any type of wireless communication mechanism that may communicate signals between devices, such as client devices or computing devices, within the network or between networks.

[0030] A computing device may be capable of transmitting or receiving signals, such as via a wired or wireless network, or may be capable of processing or storing signals as physical memory states, such as in a memory, and thus may operate as a server. Thus, devices capable of operating as a server may include, by way of example, dedicated rack-mounted servers, desktop computers, laptop computers, set-top boxes, integrated devices combining various features, such as features of two or more of the aforementioned devices, and the like. Although servers may vary widely in configuration or capabilities, in general, a server may include one or more central processing units and memory. A server may also include one or more mass storage devices, one or more power sources, one or more wired or wireless network interfaces, one or more input / output interfaces, or one or more operating systems, such as Windows Server, Mac "OS X", Unix, "Linux", FreeBSD, and the like.

[0031] In a non-limiting embodiment, the wearable device may include one or more sensors. The term "sensor" may refer to any hardware or software used to detect changes in various physical quantities caused by the activity or movement of the pet, such as actuators, gyroscopes, magnetometers, microphones, pressure sensors, or any other device that may be used to detect the displacement of an object. In one non-limiting example, the sensor may be a three-axis accelerometer. The one or more sensors or actuators may be included in a microelectromechanical system (MEMS). MEMS, also called MEMS devices, may include one or more small mechanical and / or electromechanical elements that function as sensors and / or actuators and may facilitate detecting positional changes, movement, and / or acceleration. In other embodiments, any other sensor or actuator may be used to detect any physical property, change, or quantity. The wearable device, also called a collar device, may also include one or more transducers. The transducers are used to convert the physical property, change, or quantity detected by the sensors and / or actuators into electrical signals, which may be transmitted from the one or more wearable devices to a server over a network.

[0032] FIG 1 illustrates a diagram of a system used to track and monitor a pet, according to a non-limiting embodiment. In particular, as shown in FIG 1, the system 100 may include a tracking device 102, a mobile device 104, a server 106, and / or a network 108. The tracking device 102 may be a wearable device, as shown in FIGS. 27-47. The wearable device may be located on the pet's collar and may be used to track, monitor, and / or detect the pet's activity using one or more sensors.

[0033] As shown in FIG. 1, the tracking device 102 may comprise a computing device designed to be worn or otherwise carried by a user or other entity, such as a pet or animal. The term "animal" or "pet" as used in this disclosure may refer to domestic animals, including domestic dogs, domestic cats, horses, dairy cows, ferrets, rabbits, pigs, rats, mice, gerbils, hamsters, goats, and the like. Domestic dogs and cats are specific, non-limiting examples of pets. The term "animal" or "pet" as used in this disclosure may also refer to wild animals, including, but not limited to, bison, elk, deer, deer, ducks, birds, fish, and the like.

[0034] In one non-limiting embodiment, the tracking device 102 may include the hardware shown in FIG. 2. The tracking device 102 may be configured to collect data generated by various hardware or software components, commonly referred to as sensors, within the tracking device 102. For example, a GPS receiver or one or more sensors, such as an accelerometer, gyroscope, or any other device or component, may be used to record, collect, or receive data regarding the movement or activity of the tracking device 102. In some non-limiting embodiments, the activity of the tracking device 102 may mimic the movement of the pet to which the tracking device is attached. The tracking device 102 may be attached to the pet's collar as described in U.S. Patent Application No. 14 / 231,615, the contents of which are incorporated herein by reference in their entirety, but in other embodiments, the tracking device 102 may be attached to any other object worn by the pet. In some non-limiting embodiments, the tracking device 102 may be located on or within the pet itself, such as a microchip implanted in the pet.

[0035] As described in more detail herein, the tracking device 102 may further comprise a processor capable of processing one or more data collected by the tracking device 102. The processor may be implemented by any computing or data processing device, such as a central processing unit (CPU), digital signal processor (DSP), application specific integrated circuit (ASIC), programmable logic device (PLD), field programmable gate array (FPGA), digitally enhanced processing circuit, or similar device or combination thereof. The processor may be implemented as a single controller or multiple controllers or processors. In some non-limiting embodiments, the tracking device 102 may be configured to collect, sense, or receive data, and / or pre-process data before transmission. In addition to sensing, recording, and / or processing data, the tracking device 102 may further be configured to transmit data, including location, and any other data monitored or tracked to another device or server over the network 108. In non-limiting embodiments, the tracking device 102 may continuously transmit any tracked or monitored data to the network. In other non-limiting embodiments, the tracking device 102 can transmit any tracked or monitored data discretely. Discrete transmission can be transmitting data after a finite amount of time. For example, the tracking device 102 can transmit data once an hour. This can help reduce battery power consumed by the tracking device 102 as well as conserve network resources such as bandwidth.

[0036] As shown in FIG. 1, the tracking device 102 may communicate with a network 108. Although shown as a single network, the network 108 may include multiple networks that allow communication between the devices. The network 108 may be a wireless-based communications network using any available wireless access technology. The available wireless access technologies may include, for example, Bluetooth, Wireless Local Area Network (WLAN), Global System for Mobile Communications (GMS), Universal Mobile Telecommunications System (UMTS), any Third Generation Partnership Project ("3GPP") technology (including Long Term Evolution (LTE), LTE Advanced, Third Generation Technology (3G)), or Fifth Generation (5G) / New Radio (NR) technology. The network 108 may communicate with the tracking device 102, the server 106, and / or the mobile device 104 using any of the above wireless access technologies, or any other available wireless access technology.

[0037] In one non-limiting embodiment, the network 108 may include a WLAN, such as a Wireless Fidelity (Wi-Fi) network defined by the IEEE 802.11 standard or an equivalent. In this embodiment, the network 108 may enable the transfer of location and / or any tracked or monitored data from the tracking device 102 to the server 106. The network 108 may also enable the transfer of data between the tracking device 102 and the mobile device 104. In another embodiment, the network 108 may include a mobile phone network, such as a cellular network. In this embodiment, data may be transferred between devices similarly to the embodiment in which the network 108 is a WLAN. In a non-limiting embodiment, the tracking device 102, also referred to as a wearable device, transmits data to the server 106 primarily or only when connected to a WLAN network, thereby reducing network bandwidth and increasing battery life. When not connected to a WLAN, the tracking device 102 may enter a power saving mode in which it may still monitor and / or track data but may not transfer any collected data to the server 106. This may also help extend the battery life of the tracking device 102.

[0038] In one non-limiting embodiment, the tracking device 102 and the mobile device 104 can transfer data directly between the devices. Such direct transfer may be referred to as device-to-device or mobile device-to-mobile device communication. Although described separately, the network 108 may include multiple networks. For example, the network 108 may include a Bluetooth network, a wireless local area network, and a cellular network, which may facilitate enabling data transfer between the tracking device 102 and the mobile device 104.

[0039] The system 100 may further include a mobile device 104. The mobile device 104 may be any available user device or mobile station, such as a mobile phone, a smart phone, or a multimedia device, or a tablet device. In other embodiments, the mobile device 104 may be a computer, such as a laptop computer with wireless communication capabilities, a personal digital assistant (PDA) with wireless communication capabilities, a portable media player, a digital camera, a portable video camera, a navigation device with wireless communication capabilities, or any combination thereof. As previously described, the mobile device 104 may communicate with the tracking device 102. In these embodiments, the mobile device 104 may receive location, data about the pet, health assessments, and / or health recommendations from the tracking device 102, the server 106, and / or the network 108. The tracking device 102 may also receive data from the mobile device 104, the server 106, and / or the network 108. In one non-limiting embodiment, the tracking device 102 may receive data regarding the proximity of the mobile device 104 to the tracking device 102 or the identity of a user associated with the mobile device 104. A user associated with the mobile device 104 may be, for example, a pet owner.

[0040] The mobile device 104 (or non-mobile device) may also communicate with and receive data from the server 106. For example, the server 106 may include one or more application servers providing network-connected applications or application programming interfaces (APIs). In one non-limiting embodiment, the mobile device 104 may include one or more mobile or web-based applications that communicate with the server 106 via the APIs to retrieve data and present it within the applications. In one non-limiting embodiment, the server 106 may provide a visualization or display of the location or data received from the tracking device 102. For example, the visualized data may include graphs, charts, or other representations of the data received from the tracking device 102.

[0041] 2 illustrates a device that may be used to track and monitor a pet, according to a non-limiting embodiment. Device 200 may be, for example, tracking device 102, server 106, or mobile device 104. Device 200 includes a CPU 202, memory 204, non-volatile storage 206, sensors 208, a GPS receiver 210, a cellular transceiver 212, a Bluetooth transceiver 216, and a wireless transceiver 214. The device may include any other hardware, software, processors, memory, transceivers, and / or graphical user interfaces.

[0042] As discussed with respect to FIG. 2, device 200 may be a wearable device designed to be worn or otherwise carried by a pet. Device 200 includes one or more sensors 208, such as a three-axis accelerometer. The one or more sensors may be used in combination with, for example, a GPS receiver 210. GPS receiver 210 may be used with sensor 208 to monitor device 200 and determine its location (via GPS receiver 210) and its acceleration (via sensor 208). Although shown as single components, sensor 208 and GPS receiver 210 may alternatively include multiple components each providing similar functionality. In a non-limiting embodiment, GPS receiver 210 may be a Global Navigation Satellite System (GLONASS) receiver.

[0043] The sensors 208 and GPS receiver 210 generate and transmit data to other components via the CPU 202, as described in more detail herein. Alternatively or additionally, the sensors 208 and GPS receiver 210 may transmit data to the memory 204 for short-term storage. In one non-limiting embodiment, the memory 204 may comprise a random access memory device or similar volatile storage device. The memory 204 may be any suitable storage device, such as, for example, a persistent computer-readable medium, a hard disk drive (HDD), a random access memory (RAM), flash memory, or other suitable memory.

[0044] Alternatively or additionally, the sensors 208 and GPS receiver 210 may transmit data directly to the non-volatile storage 206. In this embodiment, the CPU 202 can access the data in the memory 204 (e.g., location and / or event data). In some non-limiting embodiments, the non-volatile storage 206 may comprise solid-state storage (e.g., flash storage) or conventional storage (e.g., hard disk). In particular, the GPS receiver 210 may transmit location data (e.g., latitude, longitude, etc.) to the CPU 202, memory 204, or non-volatile storage 206 in a similar manner. In some non-limiting embodiments, the CPU 202 may comprise a field programmable gate array or a specialized application specific integrated circuit.

[0045] As shown in Fig. 2, the device 200 includes multiple network interfaces including a cellular transceiver 212, a wireless transceiver 214, and a Bluetooth transceiver 216. The cellular transceiver 212 allows the device 200 to transmit data processed by the CPU 202 to a server via any wireless access network. The CPU 202 may also determine the type and content of data to be transferred using the cellular transceiver 212, the wireless transceiver 214, and the Bluetooth transceiver 216 based on detected network conditions. The transceivers 212, 214, 216 may each independently be a transmitter, a receiver, or both a transmitter and a receiver, or a unit or device that may be configured for both transmission and reception. The transmitter and / or receiver (as far as the radio part is concerned) may also be realized as a remote radio head located not within the device itself, but for example in a mast.

[0046] FIG. 3 is a logical block diagram illustrating a device that may be used to track and monitor a pet, according to a non-limiting embodiment. As shown in FIG. 3, a device 300, such as the tracking device 102 shown in FIG. 1 or the portable device 104 shown in FIG. 1, also referred to as a wearable device, may include a GPS receiver 302, a geofence detector 304, a sensor 306, a storage device 308, a CPU 310, and a network interface 312. A geofence may refer to a geolocation fence, as described below. The GPS receiver 302, the sensor 306, the storage device 308, and the CPU 310 may be similar to the GPS receiver 210, the sensor 208, the memory 204 / non-volatile storage device 206, and the CPU 202, respectively. The network interface 312 may correspond to one or more of the transceivers 212, 214, 216. The device 300 may also include one or more power sources, such as a battery. The device 300 may also include a charging port, which may be used to charge the battery. The charging port may be, for example, a Type A Universal Serial Bus (USB) port, a Type B USB port, a mini USB port, a micro USB port, or any other type of port. In other non-limiting embodiments, the battery of device 300 may be charged wirelessly.

[0047] In this embodiment, the GPS receiver 302 records position data associated with the device 300. The position data includes a number of data points that represent the position of the device 300 as a function of time.

[0048] In one non-limiting embodiment, the geofence detector 304 stores details about known geofence zones. For example, the geofence detector 304 may store multiple latitude and longitude points for multiple polygonal geofences. The latitude and / or longitude points or coordinates may be manually entered by a user and / or automatically detected by the wearable device. In other embodiments, the geofence detector 304 may store names of known WLAN network service set identifiers (SSIDs) and associate each SSID with a geofence (described in more detail with respect to FIG. 4). In one non-limiting embodiment, the geofence detector 304 may store one or more thresholds in addition to the SSIDs for determining when the device 300 has left a geofence zone. Although shown as a separate component, in some non-limiting embodiments, the geofence detector 304 may be implemented within the CPU 310, for example as a software module.

[0049] In one non-limiting embodiment, the GPS receiver 302 may transmit latitude and longitude data to the geofence detector 304 via the storage device 308 or indirectly to the storage device 308 via the CPU 310. A geofence may be a virtual fence or safe space for a particular pet. A geofence may be defined based on latitude and / or longitude coordinates and / or by the boundaries of a particular WLAN connection signal. For example, the geofence detector 304 may receive latitude and longitude data representing the current location of the device 300 and determine whether the device 300 is within or has left a geofence zone. If the geofence detector 304 determines that the device 300 has left a geofence zone, the geofence detector 304 may transmit a notification to the CPU 310 for further processing. After the notification is processed by the CPU 310, the notification may be transmitted to the mobile device directly or via a server.

[0050] Alternatively, the geofence detector 304 may query the network interface 312 to determine whether the device is connected to a WLAN network. In this embodiment, the geofence detector 304 may compare the current WLAN SSID (or no SSID) to a list of known SSIDs. The list of known SSIDs may be based on WLAN connections previously approved by the user. The user may be prompted to approve the SSID during the setup process for the particular wearable device, for example. In another example, the list of known SSIDs may be automatically created based on WLAN connections known to the user's mobile device. If the geofence detector 304 does not detect that the device 300 is currently connected to a known SSID, the geofence detector 304 may send a notification to the CPU 310 that the device has left the geofence zone. Alternatively, the geofence detector 304 may receive the strength of the WLAN network and determine whether the current strength of the WLAN connection is within a predefined threshold. If the WLAN connection is outside the predefined threshold, the wearable device may be approaching the outer edge of the geofence. Upon receiving a notification that a network strength threshold has been exceeded, the user may receive advance warning that the pet is about to leave the geofence.

[0051] As shown in FIG. 3, the device 300 further includes a storage device 308. In one non-limiting embodiment, the storage device 308 may store past data sensed or received by the device 300. For example, the storage device 308 may store past location data. In another non-limiting embodiment, instead of storing previously sensed and / or received data, the device 300 may transmit the data to a server, such as the server 106 shown in FIG. 1. The past data may then be used to determine a health indicator. The health indicator may be stored in the server. The server may compare the determined health indicator based on the recently received data to the stored health indicator (based on previously stored data). Alternatively, in a non-limiting embodiment, the device 300 may use its own computing power or hardware to determine the health indicator. Tracking changes in the health indicator or metric using the device 300 may facilitate limiting or avoiding transmission of data to the server. The health assessment and / or health recommendation made by the server 106 may be based on previously stored data. The health assessment may include, for example, a skin disease diagnosis such as a flare-up, an ear infection, an arthritis diagnosis, a heart attack, and / or a pancreatic attack.

[0052] In one non-limiting embodiment, the stored data may include data describing the walking environment in detail, including the time the tracking device last left a geofenced zone, the location of the tracking device, and motion data associated with the device (e.g., speed, acceleration, etc.). The time may be determined by a time stamp received from a GPS receiver or an internal timer in the tracking device.

[0053] The CPU 310 can control access to the storage device 308, retrieve data from the storage device 308, and transmit data to a network-connected device via the network interface 312. As described more fully with respect to FIG. 4, the CPU 310 can receive indications of geofence zone exits from the geofence detector 304 and communicate with the mobile device using the network interface 312. In one non-limiting embodiment, storing the location data can include associating a time stamp with the data. In some non-limiting embodiments, the CPU 310 can retrieve location data from the GPS receiver 302 at a predetermined time interval. For example, the predetermined time interval can be once every three minutes. In some non-limiting embodiments, the time interval can be dynamically changed based on an estimated walking distance or remaining battery life of the device 300. The CPU 310 can also transmit the location data to a remote device via the network interface 312.

[0054] FIG. 4 is a flow diagram illustrating a method for tracking a pet, according to a non-limiting embodiment. In step 402, method 400 may be used to monitor the location of the device. In one non-limiting embodiment, monitoring the location of the device may include discretely monitoring the GPS location of the device, i.e., at regular intervals. For example, in step 402, the wearable device may discretely poll the GPS receiver every 5 seconds to retrieve the latitude and longitude of the device. Alternatively, in other non-limiting embodiments, continuous polling of the GPS location may be used. Discretely polling the GPS receiver, as opposed to continuously polling the device, may extend the battery life of the portable device and reduce the number of network or device resources consumed by the portable device.

[0055] In other non-limiting embodiments, method 400 may utilize other methods for estimating the device's location without relying on the device's GPS location. For example, method 400 may monitor the device's location by determining if the device is connected to a known WLAN connection and using the connection to the WLAN as an estimate of the device's location. In yet another non-limiting embodiment, the wearable device may be paired with a portable device via a Bluetooth network. In this embodiment, method 400 may query the paired device to determine its location, for example, using the GPS coordinates of the portable device.

[0056] In step 404, the method 400 may include determining whether the device has left a geofenced zone. As described above, in one non-limiting embodiment, the method 400 may include continuously polling a GPS receiver to determine the latitude and longitude of the device. In this embodiment, the method 400 may compare the received latitude and longitude to a known geofenced zone. A geofenced area includes a set of latitude and longitude points that define an area, such as a polygonal area. In the case of using a WLAN to indicate location, the method 400 may determine that the device has left a geofenced zone when the presence of a known WLAN is not detected. For example, the tracking device may be configured to identify a home network (e.g., using the SSID of the network). If the device is present in the home (e.g., if a pet is present in the home), the method 400 may determine that the device has not left a geofenced zone. However, when the device moves outside of the range of a known WLAN, method 400 may determine that the pet has left the geofence zone, and thus implicitly delineate the geofence zone based on the contours of the WLAN signal.

[0057] Alternatively or additionally, method 400 may use a continuous detection method to determine if the device has left the geofence zone. Specifically, WLAN networks typically experience degraded signal strength as the receiver moves farther away from a wireless access point or base station. In one non-limiting embodiment, method 400 may receive known WLAN signal strengths from a wireless transceiver. In this embodiment, method 400 may set one or more predefined thresholds to determine if the device has left the geofence.

[0058] For example, a hypothetical WLAN may have a signal strength between 10 and zero, representing the strongest possible signal and no detected signal, respectively. In non-limiting embodiments, method 400 may monitor for a signal strength of zero to determine that the device has left the geofence. Alternatively or additionally, method 400 may set a threshold signal strength value of three as the boundary of the geofence region. In this example, method 400 may determine that the device has left the geofence when the network's signal strength drops below a value of three. In some non-limiting embodiments, method 400 may utilize a timer to account for the likelihood that the network signal strength will return above a predetermined threshold. In this example, method 400 may account for temporary interruptions in WLAN signal strength to avoid false positives and / or short exits.

[0059] In method 400, if the server determines that the wearable device has not left the geofence zone, method 400 may continue to monitor the device location discretely or continuously at step 402. Alternatively, if method 400 determines that the device has left the geofence, the sensor may send a signal to the wearable device to instruct the wearable device to turn on a lighting device, as shown in step 406. The lighting device may include, for example, a light emitting diode (LED) or any other light. The lighting device may be located within the housing of the wearable device and may illuminate at least the top cover of the wearable device (also referred to as the wearable device). In yet another example, the lighting device may illuminate at least a portion and / or the entire surface of the wearable device. In a non-limiting embodiment, instead of a lighting device, the wearable device may include any other indicator, for example, an audio device. The audio device may include a speaker and / or a vibration device. At step 406, any of the above indicators, the lighting device, the audio device, or the vibration device may be turned on or activated.

[0060] In a non-limiting embodiment, the mobile device user may be prompted to determine whether the wearable device has left the geofenced zone. For example, the wearable device may be paired with the mobile device via a Bluetooth connection. In this embodiment, method 400 may include alerting the device via the Bluetooth connection that the lighting device has been turned on at step 406 and / or that the wearable device has left the geofenced zone at step 404. The user may be able to determine that the wearable device has left the geofenced zone (e.g., by providing an on-screen notification). Alternatively, the user may be notified by receiving a notification from a server based on data received from the mobile device.

[0061] Alternatively or additionally, method 400 can infer the start of a walk based on the time of day. For example, a user may schedule a walk for a certain time of day (e.g., morning, afternoon, or evening). To detect if the device has left the geofenced zone, method 400 may further check known scheduled walks to determine if the timing of the geofence exit occurred at an expected walk time (or within an acceptable deviation). If the timing indicates an expected walk time, notification to the user that the wearable device has left the geofenced zone may be omitted.

[0062] Alternatively or additionally, method 400 may use machine learning techniques to infer the start of a walk without requiring such input from the user. Machine learning techniques, such as feed-forward networks, deep feed-forward networks, deep convolutional networks, and / or long-term or short-term memory networks, may be used on any data received by the server and sensed by the wearable device. For example, during the first few times that method 400 detects that the wearable device has left the geofence zone, method 400 may continue to prompt the user to confirm that the user knows the location of the wearable device. Once method 400 receives a confirmation or denial from the user, method 400 may train a learning machine located within the server to identify conditions associated with leaving the geofence zone. For example, after several quick confirmations, the server may determine that the tracking device repeatedly leaves the geofence zone between 7:00 a.m. and 7:30 a.m. on weekdays (i.e., confirming a morning walk for the pet). In this regard, the server may learn that the same event (e.g., a morning walk) may occur later on the weekend (e.g., between 8:00 a.m. and 8:30 a.m.). Thus, the server may train itself to identify various times when the wearable device leaves the geofenced zone and not react to such exits, for example, if an exit is detected between 8:00 a.m. and 8:30 a.m. on a weekend, the server will not instruct the wearable device to turn on the light device 406.

[0063] In a non-limiting embodiment, the wearable device and / or server may continue to monitor the location of the wearable device and record its GPS location, as shown in step 408. In step 410, the wearable device may transmit location details to the server and / or mobile device.

[0064] In one non-limiting embodiment, the method 400 can continuously poll the GPS location of the wearable device. In some non-limiting embodiments, the polling interval of the GPS device can be adjusted based on the battery level of the device. For example, the polling interval can be reduced if the battery level of the wearable device is low. In one non-limiting example, the polling interval can be reduced from every 3 minutes to every 15 minutes. In other embodiments, the polling interval can be adjusted based on the expected length of time of the wearable device outside the geofenced zone. That is, if the time outside the geofenced zone is expected to last 30 minutes (e.g., while walking a dog), the server and / or the wearable device can calculate the optimal polling interval based on the battery life. As described above, the length of the walk can be manually entered by the user or determined using machine learning or artificial intelligence algorithms based on previous walks.

[0065] In step 412, the server and / or wearable device may determine whether the wearable device has entered a geofenced zone. If not, steps 408, 410 may be repeated. Entering a geofenced zone may be a re-entry into the geofenced zone. That is, the wearable device may be determined to have entered a geofenced zone after having left the geofenced zone. As discussed above, the server and / or wearable device may utilize a polling interval to determine how frequently to transmit data. In one non-limiting embodiment, the server and / or wearable device may transmit location data using a cellular or other wireless network. Methods for transmitting location data over a cellular network are more fully described in commonly owned U.S. Patent Application No. 15 / 287,544, which is incorporated herein by reference in its entirety.

[0066] Finally, once the server and / or wearable device determines that the wearable device has entered a geofenced zone, the lighting or any other indicators on the wearable device may be turned off. In some non-limiting embodiments, not shown in FIG. 4, a user may choose to turn off the lighting when the wearable device leaves the geofenced zone. For example, once a user of a portable device determines that the wearable device has left the geofenced zone, the user may instruct the wearable device to instruct the server to turn off the lighting, or may directly instruct the wearable device to turn off the lighting.

[0067] FIG. 5 is a flow diagram illustrating a method for tracking and monitoring a pet, according to non-limiting embodiments. The steps of the method shown in FIG. 5 may be performed by a server, a wearable device, and / or a mobile device. The wearable device may sense, detect, or collect data related to the pet. The data may include, for example, data related to the pet's location or movement. In a non-limiting example, the wearable device may include one or more sensors that enable the wearable device to detect the movement of the pet. In some non-limiting embodiments, the sensor may be a three-axis accelerometer mounted on a collar that may enable the wearable device to detect various body movements of the pet. The various body movements may include, for example, any body movements associated with itching, scratching, licking, walking, drinking, eating, sleeping, and shaking, and / or any other body movements associated with the pet's movements. In some examples, the one or more sensors may detect the pet jumping around, getting excited for food, gobbling, drinking from a bowl on the wall, and / or walking around the room. The one or more sensors may also detect activity of a pet following a treatment or veterinary visit (such as a neuter or spay-hysterectomy visit).

[0068] In non-limiting embodiments, the data collected by the one or more sensors may be combined with data collected by other sensors. In one non-limiting example, the data collected by the one or more sensors may be combined with video and / or audio data captured using a video recorder. Combining data from one or more sensors with data from a video recorder may be referred to as data preparation. During data preparation, the video and / or audio data may utilize video labeling, e.g., behavioral labeling software. The video and / or audio data may be synchronized and / or stored together with the data collected from the one or more sensors. Synchronization may include comparing the sensor data with the video labels and aligning the sensor data with the video labels with minute, second, or sub-second accuracy. The data may be aligned manually by a user or automatically, e.g., using a semi-supervised approach to estimate the misalignment. The combined data from one or more sensors and from the video recorder may be analyzed using machine learning or any of the algorithms described herein. The data may also be labeled as training data, validation data, and / or test data.

[0069] The data may be sensed, detected, or collected continuously or discretely as described in FIG. 4 for location data. In a non-limiting embodiment, the pet's activity may be sensed or detected continuously by the wearable device and the data may be collected continuously, but the wearable device may transmit the information to the server discretely to conserve battery power and / or network resources. In other words, the wearable device may continuously monitor or track the pet, but transmit the collected data every finite amount of time. The finite amount of time used for transmission may be, for example, one hour.

[0070] In step 501, data regarding the pet from the wearable device may be received at a server and / or at the user's mobile device. Once received, the data may be processed by the server and / or the mobile device to calculate one or more health indicators for the pet, as shown in step 502. The server may utilize machine learning tools, such as deep neural networks using convolutional and / or recurrent neural network layers as described below. The machine learning tools may be referred to as activity recognition algorithms or models. In a non-limiting embodiment, the machine learning tools may include one or more layer modules as shown in FIG. 7. Using the machine learning tools, health indicators, also referred to as behaviors, of the pet wearing the device may be calculated. The one or more health indicators may consist of metrics for itching, scratching, licking, walking, drinking, eating, sleeping, and shaking. The metrics may be, for example, the distance walked by the pet, the amount of sleep, and / or the amount of itching.

[0071] The machine learning tool may be trained. To train the machine learning tool, for example, a server may collect data from multiple wearable devices. Collecting data from multiple wearable devices may be referred to as crowdsourcing data. The collected data from one or more pets may be integrated and / or categorized to learn one or more trends or relationships present in the data. The learned trends or relationships may be used by the server to calculate, predict, and / or estimate a health indicator from the received data. The health indicator may be used to determine any behaviors exhibited by the pet that may affect the pet's health. Machine learning may also be used to model the relationship between the health indicator and a possible impact on the pet's health. For example, the likelihood that the pet may suffer from a disease or set of diseases, such as a skin disease. The machine learning tool may be automated or semi-automated. In a semi-automated model, the machine learning may be assisted by a human programmer who intervenes in the automated process and helps identify or validate one or more trends or models present in the data being processed during the machine learning process.

[0072] In a non-limiting embodiment, the machine learning tool used to convert data such as time series accelerometer readings into predictive health indicators can use a windowing method to predict behavior during a small time window. Such an embodiment may generate one prediction per window. However, other non-limiting embodiments do not use small time windows and data therebetween, and the machine learning tool can handle the total amount of data. The data received from the wearable device can be aggregated before being fed to the machine learning tool, thereby allowing analysis of a large number of data points. For example, the aggregation of the data can divide the data points originally received in a 3 Hertz frequency window into minutes of an hour, hours of a day, months of a year, or any other period that can facilitate processing and aid the machine learning tool in modeling. If the data is aggregated more than once, a hierarchy can be established after the data aggregation. The hierarchy can be based on the periodicity of the data bins into which the aggregated data falls, with each re-aggregation of the data reducing the number of bins into which the data can fall.

[0073] For example, 720 data points (in some non-limiting embodiments, processed individually using smaller time windows) may be aggregated into 10 data points for processing by the machine learning tool. In another example, the aggregated data may be re-aggregated into a smaller number of bins to help further reduce the number of data points processed by the machine learning tool. Working with the total amount of data may help generate a larger number of matches and / or predictions. Other non-limiting embodiments may learn and model trends more efficiently to reduce the amount of time required for processing and improve accuracy. The above aggregation hierarchies may also help reduce the amount of storage. Instead of storing raw data or data lower in the aggregation hierarchy, non-limiting embodiments may store the data in a highly aggregated hierarchical format.

[0074] In some other embodiments, aggregation occurs after the machine learning process using neural networks, and the data may simply be resampled, filtered, and / or transformed before being processed by the machine learning tool. Filtering may include removing interference such as brown or white noise. Resampling may include stretching or compressing the data, and transforming may include flipping the axes of the received data. Transformations may also take advantage of natural symmetries, such as bilateral symmetry, and different collar positions of the data signal. In some non-limiting embodiments, data augmentation may include adding noise, such as brown, pink, or white noise, to the signal.

[0075] In step 503, a health assessment of the pet based on one or more health indicators may be performed. The health assessment may include, for example, an indication of one or more illnesses, health abnormalities, and / or any combination thereof as determined and / or presented by the health indicator. The health abnormalities may include, for example, one or more of skin disease, ear infection, arthritis, heart attack, tooth fracture, cruciate ligament rupture, pancreatic attack, and / or any combination thereof. In a non-limiting embodiment, the server may instruct the wearable device to turn on a lighting device based on the health assessment of the pet as shown in step 504. In step 505, the health indicator may be compared to one or more stored health indicators (which may be based on previously received data). If different thresholds are detected by comparing the health indicator to the stored health indicators, the health assessment may reflect such detection. For example, the server may detect that the pet sleeps less compared to a particular threshold, is more itchy compared to a particular threshold, or eats less compared to a particular threshold. A health assessment may be performed based on these specified or pre-set thresholds. In some non-limiting embodiments, the thresholds may also be determined using the machine learning tools described above. A health assessment may, for example, identify that a pet is overweight or that a pet may be potentially ill.

[0076] In step 506, the server can determine a health or fitness prompt for the pet based on the health assessment. In a non-limiting embodiment, the fitness prompt can be an exercise regimen for the pet. For example, the fitness prompt can be to have the pet walk a certain number of steps per day and / or run a certain number of steps per day. The health or fitness prompt can, for example, provide the user with a recommendation to treat a possible health risk for the pet. The health recommendation can, for example, inform the user of the health assessment and recommend that the user take the pet to a veterinarian for diagnosis and / or treatment, or provide a specific treatment recommendation, for example, a recommendation to feed the pet a certain food or administer an over-the-counter medication. In other non-limiting embodiments, the health recommendation can include a recommendation to purchase one or more pet foods, one or more pet products, and / or any combination thereof. In steps 507 and 508, the health assessment, health recommendations, fitness prompts, and / or any combination thereof are transmitted from the server to the mobile device, where the health assessment, health recommendations, and / or fitness prompts may be displayed, for example, on a graphical user interface of the mobile device.

[0077] In some non-limiting embodiments, the data received by the server may include location information measured or obtained using GPS. The data may be received at the wearable device via GPS and transmitted to the server. The location data may be used, as well as any other data described above, to calculate one or more health indicators for the pet. In non-limiting embodiments, monitoring the location of the wearable device may include identifying an active wireless network within a perimeter of the wearable device. If a wireless network is within a perimeter of the wearable device, the wearable device may connect to the wireless network. When the wearable device leaves the geofence zone, there may no longer be an active wireless network within the perimeter of the wearable device. In other embodiments, the geofence may be pre-defined using latitude and longitude coordinates.

[0078] A non-limiting embodiment may be directed to a method of analyzing data. The method may include receiving data at a device. The data may include at least one of financial data, cybersecurity data, electronic health records, acoustic data, human activity data, or pet activity data. The method may also include analyzing the data with two or more layer modules. Each layer module may include at least one of a many-to-many approach, striding, downsampling, pooling, multi-scaling, or batch normalization. The method may also include determining an output based on the analyzed data. The output may include a health assessment, a health recommendation, a financial prediction, or a security recommendation. The two or more layers may include at least one of a full resolution convolutional neural network, a first pooling stack, a second pooling stack, a resampling step, a bottleneck layer, a recurrent stack, or an output module. In some embodiments, the determined output may be displayed on the mobile device.

[0079] The data may be received, processed, and / or analyzed as described in the embodiment shown in FIG. 5. In a non-limiting embodiment, the data may be processed using a time series classification algorithm. A time series classification algorithm may be used to evaluate or predict data over a particular period of time. An activity recognition algorithm that tracks a pet's momentary activity over time may be an example of a time series classification algorithm. Some time series classification algorithms may utilize K-nearest neighbor and support vector machine techniques, while other algorithms may utilize deep learning based techniques, such as the examples below.

[0080] In a non-limiting embodiment, the activity recognition algorithm may utilize a machine learning model. In such an embodiment, an appropriate time series may be obtained and used to frame split the received data. Then, hand-crafted statistical and / or spectral feature vectors may be calculated over one or more finite time windows. Features may be individual measurable characteristics observed via the wearable device. A feature vector may include a set of one or more features. Hand-crafted may refer to a feature vector obtained using a human-defined algorithm. A training model, such as K-nearest neighbor (KNN), Naive Bayes (NB), decision tree or random forest, support vector machine (SVM), or any other known training model, may map the calculated feature vector to an activity prediction. The training model may be evaluated with new or provided time series data to infer activity.

[0081] One or more training models may be used or combined to improve the prediction results. For example, ensemble-based methods may be used to combine one or more training models. The Collection of Transform-Based Ensembles (COTE) and the Hierarchical Voting variant HIVE-COTE are examples of ensemble-based methods.

[0082] Rather than using machine learning models or tools such as KNN, NB, or SVM, some other embodiments may utilize one or more deep learning or neural network models. Deep learning or neural network models do not rely on hand-crafted feature vectors. Instead, deep learning or neural network models use learned feature vectors obtained from a training procedure. In non-limiting embodiments, a neural network may include a computational graph consisting of a number of primitive blocks, each block performing a weighted sum of its inputs to introduce non-linearity. In some non-limiting embodiments, a deep learning activity recognition model may include a convolutional neural network (CNN) as a component. In some examples, a CNN may convolve a trainable fixed-length kernel or filter along its inputs, while a neural network trains learned weights for every input-output pair. In other words, a CNN may learn to recognize small primitive features (low level) and combine them into complex (high level) combinations.

[0083] In non-limiting embodiments, pooling, padding, and / or striding may be used to reduce the size of the CNN output to the dimensions over which the convolutions are performed, thereby reducing computational cost and / or making it less susceptible to overtraining. Striding may describe the step size or number of steps over which the filter window slides, while padding may include padding regions of data with zeros and buffering the data before or after striding. Pooling may include, for example, simplifying the information collected by a convolutional layer or any other layer to create a condensed version of the information within those layers. In some examples, a one-dimensional (1D) CNN may be used to process fixed-length time series segments generated with a sliding window. Such a 1D CNN may be implemented in a many-to-one configuration utilizing pooling and striding, and the output of the final CNN layer may be concatenated. A fully connected layer may be used to generate class predictions at one or more time steps.

[0084] In contrast to 1D CNNs, which convolve a fixed-length kernel along the input signal, recurrent neural networks (RNNs) process each time step sequentially, such that the final output of the RNN layer is a function of all previous time steps. In a non-limiting embodiment, a variant of RNN known as a long-short-term memory (LSTM) model can be used. LSTMs contain memory cells and / or one or more control gates and can model time dependencies in long sequences. In some embodiments, the LSTM model is unidirectional, meaning that the model processes the time series in the order in which it was recorded or received. In another embodiment, two parallel LSTM models can be evaluated in both directions in time, forward and backward, if the entire input sequence is available. The results of two parallel LSTM models can be concatenated to form a bidirectional LSTM (bi-LSTM) that can model time dependencies in both directions.

[0085] In some non-limiting embodiments, one or more CNN models and one or more LSTM models may be combined. The combined model may include a stack of four unstrided CNN layers, possibly followed by two LSTM layers and a softmax classifier. The softmax classifier may normalize a probability distribution that includes multiple probabilities proportional to an exponential function of the input. The input signal to the CNN is, for example, not zero-padded, so that the CNN layers are not strided, but each CNN layer shortens the time series by a few samples. Since the LSTM layers are unidirectional, a softmax classifier corresponding to the final LSTM output may be used for training and evaluation and for reassembling the output time series from the sliding window segments. However, the combined model may operate in a many-to-one configuration.

[0086] FIG. 6 illustrates two example deep learning models according to non-limiting embodiments. In particular, FIG. 6 illustrates many-to-one model 601 and many-to-many model 602. In the many-to-one approach or model 601, the input is first split into fixed-length overlapping windows. The model then processes each window separately and generates a class prediction for each window, which may be concatenated into an output time series. Thus, the many-to-one model 601 may be evaluated once per window. On the other hand, in the many-to-many model 602, the entire output time series may be generated in a single model evaluation. The many-to-many model 602 may be used to process more than one input signal at a time without the need for sliding fixed-length windows.

[0087] In non-limiting embodiments, the model may incorporate features or elements taken from one or more models or techniques. Doing so helps improve the accuracy of the model, prevents bias, improves generalization, and allows faster processing of data. Using elements from a many-to-many technique may allow processing of the entire input signal, which may include one or more signals. In some non-limiting embodiments, the model may also include striding or downsampling. Each layer of the model may use striding to reduce the number of samples output after processing. Using striding or downsampling may help improve computational efficiency and allow later layers to model dynamics over a longer time range. In non-limiting embodiments, the model may also utilize multi-scaling. Multi-scaling may help downsample beyond the output frequency and model longer range time dynamics.

[0088] Models utilizing features or elements of many-to-many models, striding or downsampling, autoscaling, and multiscaling allow themselves to be applied to time series of any length. For example, the model can infer an output time series of length proportional to the input length. Using features or elements of many-to-many models, which may be called column-by-column models, may allow the model to be agnostic to the length of the input. Also, in some embodiments, larger models may not be necessary for larger time series lengths or sliding window lengths.

[0089] In a non-limiting embodiment, the model may include a stack of parameterized modules. These may be referred to as flexible layer modules (FLMs). One or more FLMs may be coupled to a signal processing stack and may be tuned and reconfigured to be efficiently trained. Each FLM may be range-preserving, meaning that the input and / or output of the FLM may have different sequence lengths due to the stride ratio and / or the period covered by the input and output may be the same. An FLM may be represented using the following notation:

[0090]

number

[0091] type represents the type of the main trainable sublayer (cnn is 1D CNN; lstm is bidirectional LSTM), and w out is the number of output channels (for CNN, the number of filters; for LSTM, the number of dimensions of the hidden state), s is the stride ratio (default 1), k is the kernel length (for CNN, default 5), and p drop may represent the dropout probability (default 0.0). In a non-limiting embodiment, 1D average pooling with stride s and pooling kernel length s reduces the output length by a factor of s when s>1.

[0092] Each FLM randomly selects a sensor channel with probability p drop A decimation layer may be attached to a 1D CNN layer or a bidirectional LSTM layer. A 1D average pooling layer pools and strides the outputs of the CNN or LSTM layers whenever s is nonzero. The 1D average pooling layer may be called a strided layer and may include a consensus pooling step so that all CNN or LSTM output samples are represented in the FLM output. Bulk normalization (b BN ) layers may also be included in the FLM layer. The batch and / or decidual layers may serve to regularize the network and improve training dynamics.

[0093] In a non-limiting embodiment, the CNN layer is configured such that the input and output signal lengths are equal.

[0094]

number

[0095] Therefore, each FLM can be constructed to zero-pad the input by in , L in ] input tensor Xin of size [w out , L out =L in / s] to an output tensor Xout. In some non-limiting embodiments, other adjustments may be added, such as one or more gated recurrent unit (GRU) layers, which may include a gating mechanism in the recurrent neural network. Other adjustments may include grouping of CNN filters, and different strategies for pooling, striding, and / or augmentation.

[0096] 7(a) and 7(b) illustrate model architectures according to non-limiting embodiments. In particular, FIG. 7(a) illustrates an embodiment of a model including one or more stacks of FLMs, such as the dropout layer 701, 1D average pooling 702, and 1D batch normalization 703 described above. FIG. 7(b) illustrates a component architecture in which one or more FLMs may be grouped into components that may be parameterized and combined to achieve time series classifiers of different speed and complexity. The component architecture may include one or more components or FLMs, such as a full resolution CNN 711, a first pooling stack 712, a second pooling stack 713, a bottleneck layer 715, a regression stack 716, and / or an output module 717. The component architecture may also include a resampling step 714. Any one or more of the components or FLMs illustrated in FIG. 7(b) may be removed.

[0097] The full resolution CNN 711 may include high resolution processing characterized as s=1 and type=CNN. In a non-limiting embodiment, the full resolution CNN 711 may be a CNN filter that processes the input signal without striding or pooling to extract information at the finest time resolution available. In some non-limiting embodiments, this layer may be applied to the full resolution input signal and may be computationally intensive. The first pooling stack 712 is used to downsample the input to the output frequency and may be characterized as s>1 and type=CNN. p1 A stack 712 of CNN modules (each strided by s) scales the input signal by the sum of the ratios

[0098]

number

[0099] Downsample by n p1 may be the number of CNN modules included in the first pooling stack 712. The output length of the stack 712 is determined by the output stride ratio S out =s^n p1 and hence the output length of the network for a particular input

[0100]

number

[0101] Here, L OUT is the output length, L IN is the input length.

[0102] The second pooling stack 713 is used to further downsample the signal and can be characterized as s>1 and type=CNN. p2 This stack of modules (each strided by s) further downsamples the output of the previous layer beyond the output frequency to capture slower time dynamics. To prevent overtraining, we i =w p s 1-i (where i=1...n p2 A resampling step 714 may also be used to process the signal. In this step, the output of the second pooling stack 713 is resampled by linear interpolation to reduce the network output length L OUTThese outputs can be concatenated with the final module output of the first pooling stack 712. Without the resampling step 714, the length of the output of the second pooling stack 713 cannot match the output length and cannot be processed together in the next layer. Exposing each intermediate output of the second pooling stack 713 to the resampling step 714, as opposed to exposing only the final output of the second pooling stack 713, can help improve the training dynamics and accuracy of the model.

[0103] The model may also include a bottleneck layer 715 that can effectively reduce the width of the concatenated output from the resampling step 714. In other words, the bottleneck layer 715 may help minimize the number of learned weights required in the regression stack 716. The bottleneck layer may allow multiple channels from the second pooling stack 713 and the resampling step 714 to be concatenated without overtraining or overly slowing down the network. As a CNN with kernel length k=1, the bottleneck layer 715 may resemble a fully connected dense network that is applied independently at each time step.

[0104] The regression stack 716 may be characterized as s=1 and type=LSTM. In a non-limiting embodiment, the regression stack 716 may be characterized as n l The stack 716 allows additional capabilities that allow modeling of long range time dynamics and / or improve the output stability of the network. The output module 717 provides predictions for each output time step and may be characterized as s=1 and k=1. As with the bottleneck layer 715, the output module 717 may be implemented as a CNN with k=1. In a non-limiting embodiment, the multi-class output is calculated by dividing the layer outputs z i Let P(z) be the class probability distribution. i This can be achieved using a softmax activation function that transforms and normalizes according to

[0105]

number

[0106] One or more layers 711-717 may be independently reconfigured or removed to optimize model performance.

[0107] FIG. 8 illustrates an example of a model according to a non-limiting embodiment. In particular, FIG. 8 illustrates a variant of the model shown in FIG. 7(b). A basic LSTM (b-LSTM) model 801 may simply include a regression stack 716 and an output module 717. In other words, a b-LSTM does not downsample the network input, but instead uses one or more FLMs followed by an output module. LSTM The pooled CNN (p-CNN) model 802 may include a full resolution CNN 711, a first pooling stack 712, a regression stack 716, and an output module 717. Thus, the p-CNN model 802 may include multiple FLMs. LSTM The model 802 may be a stack of layers, where one or more layers are strided so that the output frequency is lower than the input frequency, which may improve computational efficiency and increase the time scale that the network models compared to a non-strided CNN stack.

[0108] The pooled CNN or LSTM model 803 (p-C / L) may include a full resolution CNN 711, a first pooling stack 712, a regression stack 716, and an output module 717. The p-C / L may add one or more regression layers operating at the output frequency immediately before the output module layer. The multi-scale CNN (ms-CNN) 804 may include a full resolution CNN 711, a first pooling stack 712, a second pooling stack 713, a resampling step 714, a bottleneck layer 715, and / or an output module 717. The multi-scale CNN or LSTM (ms-C / L) 805 may include a full resolution CNN 711, a first pooling stack 712, a second pooling stack 713, a resampling step 714, a bottleneck layer 715, a regression stack 716, and / or an output module 717. The ms-CNN and ms-C / L variants modify the p-CNN and p-C / L variants by adding a second pooling stack followed by a resampling layer and a bottleneck layer. The progression from p-CNN to ms-C / L has the effect of improving the ability of the variants to model long-range temporal interactions by adding additional layers of striding and pooling and a recurrent LSTM layer.

[0109] A dataset may be used to test the effectiveness of the model. For example, the opportunity activity recognition dataset may be used to test the effectiveness of the model shown in FIG. 7(b). The opportunity activity recognition dataset may include 6 hours of recording and labeling of several subjects using a diverse array of sensors, such as 7 inertial measurement units (IMUs) including accelerometers, gyroscopes, and magnetic sensors, and 12 Bluetooth accelerometers. See Daniel Roggen et al., "Collecting Complex Activity Data Sets in Highly Rich Networked Sensor Environments," Seventh International Conference on Networked Sensing Systems (INSS'10), Kassel, Germany, (2010), available at https: / / archive.ics.uci.edu / ml / datasets / opportunity+activity+recognition. This opportunity activity recognition dataset is incorporated herein in its entirety. Each subject was recorded performing a practice session of a pre-defined scripted activity and a total of five sessions in which the subject performed several daily living activities in a random order. The dataset may be provided at 30 or 50 Hz frequency. In some examples, linear interpolation can be used to fill in missing sensor data. Also, in a non-limiting embodiment, rather than rescaling and clipping all channels to the [0,1] interval with a predefined scaling, the data can be rescaled to have zero mean and unit standard deviation according to the statistics of the training set.

[0110] FIG. 7(c) illustrates a model architecture according to a non-limiting embodiment. In particular, the embodiment of FIG. 7(c) shows a CNN model that processes accelerometer data in a single shot. The first seven layers, including a fine-scale CNN 721 and a coarse-scale RNN stack, each decimate the signal by a factor of two to model increasingly long-range effects. The final four layers, including a mixed-scale final stack 723, may combine outputs from various scales to generate predictions. For example, data from the coarse-scale RNN stack 722 may be interpolated and combined at a frequency 1 / 16 of 50 Hz.

[0111] The six model variants shown in FIG. 8 may vary significantly in implementation due to the number of layers in each component and / or the configuration of each channel, e.g., the number of striding and output layers. Therefore, in the examples below, a specific reference architecture for each variant may be tested. To accurately compare the six variants, model parameters are kept constant or slightly changed. FIG. 9 shows an embodiment of the model shown in FIG. 8 and the layer module shown in FIG. 7. Specifically, the number, size, and configuration of each layer of the tested models can be seen in FIG. 9.

[0112] FIG. 10 shows an example architecture consisting of the model shown in FIG. 8 and one or more of the layer modules shown in FIG. 7. In particular, the model shows a more detailed architecture of the ms-C / L shown in FIG. 8. The region of influence (ROI) of a particular layer refers to the maximum number of input samples that can influence the computation of the output of that particular layer. The ROI can be increased by larger kernels, larger stride ratios, and / or additional layers, and represents an upper limit on the time scale that the architecture can model. In some examples, the region of influence of a bidirectional LSTM can be the entire input, so the ROI can be calculated for a CNN-type layer. The ROI can be calculated for other FLMs in the stack. CNN FLM preceded only by layers CNN (s i , k i ) ROI for layer i i can be calculated using the following formula:

[0113]

number

[0114] As mentioned above, in a non-limiting embodiment, a model such as the models shown in Figures 7(a) and 7(b) can be a many-to-many model used to process signals of any length. However, in some non-limiting embodiments, due to memory, efficiency, and latency constraints, it can be helpful to split a long input signal into overlapping segments using a sliding window of a moderate fixed length. The sliding window can have a length of, for example, 512 samples. It can be helpful to process the segments in batches of a size appropriate to the available memory and / or to reconstruct the corresponding output signal from the processed segments. In other embodiments, the signal can be manipulated to avoid edge effects at the beginning and end of each segment. For example, overlap between segments can allow these edge regions to be removed without creating gaps in the output signal. The overlap can be 50%, or any other number between 0 and 100%. In yet another example, to prevent signal discontinuities, the segments can be averaged using a weighted window, such as a Hanning window, that downweights edge regions.

[0115] In non-limiting embodiments, validation and test set performance may be calculated using both sample-based and event-based metrics. Sample-based metrics are aggregated across all class predictions and may not be affected by the order of predictions. Event-based metrics are calculated after the output is split into individual events and may be strongly affected by the order of predictions. Sample-based precision, recall, and F1-score may be calculated for each output class, including the blank class. The F1-score takes into account both precision and recall and may be calculated by the following formula:

[0116]

number

[0117] The overall model performance is, for example, the average F1 score F averaged over the non-empty classes. 1m, or the weighted F1 score across all classes, F 1w (where each class is weighted according to its sample proportion in the ground truth label set). In another embodiment, we use a non-blank weighted F1 score F that ignores the blank class. 1w,nn In the case of event-based metrics, for example, we can combine these multiple metrics into a single digit that is suitable for summarizing the overall performance of the model, the event F1 metric F 1e In some non-limiting embodiments, F 1e can be calculated using the true positives TP, false positives FP, and false negatives FN. 1e The formula for calculating can be expressed as follows:

[0118]

number

[0119] TP events are correct events, FN events are incorrect real events, and FP events can be incorrect return events. 1e To calculate , a non-limiting embodiment can simply sum the numbers of TP, FP, and FN across all classes. This score can be weighted by event length, meaning that long events can have the same impact as short events. The training speed of the model can be measured as the total time it takes to train the model on a particular computer, and the inference speed of the model can be measured as the average time it takes to classify each input signal on a computing system (representative of the system on which the model normally operates).

[0120] FIG. 11 illustrates an example of model parameters according to a non-limiting embodiment. In particular, FIG. 11 describes parameters that may be used to train a model on a GPU. For example, the parameters may include max epochs, initial learning rate, samples per batch, training window steps, optimizer, weight decay, patience, learning rate decay, and / or window length. The training and validation sets may be divided into segments using a sliding window. The window length may be an integer multiple of the model's output stride ratio to minimize implementation complexity. Since the window length may be varied in testing, the batch size may be adjusted to keep a constant or near-constant number of input samples in one batch.

[0121] In non-limiting embodiments, the validation loss may be used as the early stopping metric. However, in some non-limiting embodiments, the number of subjects and runs in the validation set may be small and the validation loss may be too noisy to be used as an early stopping metric. Instead of the validation loss, non-limiting embodiments may use an individualized stopping metric, which is more robust and penalizes performance fluctuations. The individualized stopping metric may help limit the model from stopping until model performance stabilizes. The smoothed validation metric is an exponentially weighted running average l v / F 1w,v (which has a half-life of 3 epochs). v is the validation loss, F 1w,v can be the weighted F1 score of the validation set calculated after each training epoch. As the loss and / or F1 score improve, the smooth validation metric decreases. The instability metric is also calculated over the last 5 l v / F 1w,v The smoother validation metric and the individualized stopping metric may be calculated as the standard deviation, mean, or median of the values. The smoother validation metric and the individualized stopping metric may be summed to produce a validation point metric. The model may be validated whenever the validation point metric reaches a new minimum, and / or training may be stopped without validation after a patience epoch.

[0122] FIG. 12 shows an example of a representative model training run according to a non-limiting embodiment. In particular, FIG. 12 illustrates the training history of the ms-C / L model using the parameters shown in FIG. 11. The stopping metric 1201, validation loss 1202, validation F1 (1203), and learning rate ratio 1204 are shown in FIG. 12. For example, the individual stopping metrics may be predicted to a minimum at epoch 43, while the validation loss varies to near a global minimum at epochs 15, 24, 34, 41, and 45. Training may be stopped at epoch 53, and the model at epoch 43 may be reverted and used for subsequent inference.

[0123] In non-limiting embodiments, ensembling may be performed using multiple learning algorithms. Specifically, n-fold ensembling may be performed by performing one or more of the following steps: (a) combining the training and validation sets into one contiguous set; (b) splitting the set into n disjoint folds of contiguous samples; (c) training n independent models, where the i-th model uses the i-th fold for validation and the remaining n-1 folds for training; and (d) ensembling the n models during inference by simply averaging the outputs before applying a softmax function. In some non-limiting embodiments, to improve efficiency, the evaluation and ensembling of the n models may be performed using one computational graph.

[0124] FIG. 13 illustrates the performance of example models according to a non-limiting embodiment. In particular, FIG. 13 illustrates performance metrics for b-LSTM 801, p-CNN 802, p-C / L 803, ms-CNN 804, and ms-C / L 805. FIG. 13 also illustrates the performance metrics for two variants of ms-C / L 805, such as a 4-fold ensemble of ms-C / L 806 and a quarter-reduced version 807 (w out The performance of the 4-fold ms-C / L model 806 may be more accurate than other variants. Other fold variants such as 3- to 5-fold ms-C / L ensembles may work well for many tasks and datasets, especially when inference speed and model size are not as important as other metrics.

[0125] FIG. 14 illustrates a heatmap showing the performance of a model according to a non-limiting embodiment. The heatmap shown in FIG. 14 shows the difference between model outputs, such as labels (ground-truths provided in the opportunity benchmark dataset of FIG. 14) 1401, a 4-fold ensemble of multi-scale CNN / LSTM 1402, a multi-scale CNN 1403, a baseline CNN 1404, a raw LSTM 1405, and a deep convolutional LSTM 1406. In particular, FIG. 14 illustrates the ground-truth labels 1401 and model predictions for the first half of the first run of a standard opportunity test set for several models. One or more models, such as the ms-C / L architecture, generate fewer short spurious events. This helps reduce the number of false positives and may also prevent splitting otherwise correct events. For example, in the illustrated region of interest, the event-based F 1e The metric increases from 0.85 for multi-scale CNN1403 to 0.96 for the multi-scale CNN / LSTM 4-fold ensemble1402, but the per-sample F 1w The metric increases slightly from 0.93 to 0.95. The event performance achieved by one or more of the above models may eliminate the need for further event processing and downselection.

[0126] FIG. 15 illustrates performance metrics for a model according to a non-limiting embodiment. In particular, FIG. 15 shows results calculated on the entire test set for the same model shown in FIG. 14. The event-based metrics shown in FIG. 15 are the event-based accuracy P e , recall R e , F1 score F 1e , and an event summary diagram. The event summary diagram compares the ground truth labels (actual events) with the model predictions (detected events). In a non-limiting embodiment, the correct events (C) may indicate that there is a one-to-one correspondence between actual and detected events. The event summary diagram depicts the number of actual events that were missed (D: deleted) or multiple detected (F: disjointed), as well as the detected fragments (F': disjoint) and false detections (I': insertions).

[0127] As shown in the results in Figure 15, the lower performing models p-CNN 1503, b-LSTM 1504, and / or deep convolutional LSTM 1505 show low accuracy. b-LSTM 1504 correctly detected 131 of 204 events and generated 434 spurious or scattered events. The ms-CNN model 1502 shows the effect of adding strided layers to p-CNN 1503, which increases the model's domain of influence from 61 to 765 samples. This means that the ms-CNN model 1502 can model dynamics over a 12 times longer domain of influence. The 4xms-C / L ensemble 1501 can be further improved by adding LSTM layers and making it harder for a single model to register spurious events without consent from other models in the ensemble. Deep Convolutional LSTM 1505 also includes an LSTM layer, but its ROI is limited to an input window length of 24 samples, which may be approximately 3% of the ROI of ms-C / L. In a non-limiting embodiment, the hidden state of the LSTM in one window segment cannot affect the next window segment.

[0128] FIG. 16 shows the performance of an n-fold ensemble ms-C / L model according to a non-limiting embodiment. In particular, FIG. 16 shows the performance of a sample-based F 1w 1601 and Event-Based F 1e 1602 indicates the weighted F1 metric. 1w and F 1e Both improve with the number of ensemble models, reaching a plateau between 3 and 5 folds. However, the ensemble coherence ratio 1603 decreases as the number of folds increases. The effect of ensembling models on accuracy, e.g., the F 1w 1601, Event-based F 1e1602, and guess rate 1603 are graphed in FIG. 16. As above, a model may be trained on (n-1) folds, and the remaining folds may be used for validation. Thus, in a non-limiting embodiment, a 2-fold model may have a validation set equal in size to the test set, and the training and validation sets may simply be swapped within its two sub-models. A larger n-fold model may encounter a training-validation split, which may be approximately 67%:33%, 75%:25%, and 80%:20% for 3-fold, 4-fold, and 5-fold ensembles, respectively. In some non-limiting embodiments, the event-based metrics 1602 may benefit more from ensembling than the sample-wise metrics 1601, as measured by the difference between the ensembles and the sub-models shown in FIG. 16.

[0129] FIG. 17 illustrates the effect of varying the sliding window length used in the interference step, according to a non-limiting embodiment. The models illustrated in FIG. 17 are b-LSTM 1701, p-CNN 1702, p-C / L 1703, ms-CNN 1704, and ms-C / L 1705. While one or more models may process time series of any length, in a non-limiting embodiment, the use of windows may be necessitated by efficiency and memory constraints. Also, some overlap may be used to reduce edge effects in the windows. For example, a 50% overlap may be used, weighted using a Hanning window to lightly weight edges and reduce the introduction of discontinuities between windows. The batch size may be, for example, 100 windows.

[0130] Model accuracy increases monotonically with window length, but inference rates can reach a maximum for models with LSTMs, where the efficiency of constructing and reassembling longer segments and the efficiency of some parallel execution on the GPU balance the inefficiency of sequential execution of the LSTM layer on the GPU. Windows of 256 to 2048 samples tend to work well, although the balance can vary. On a CPU, shorter windows can exhibit overhead, but these effects can be less noticeable due to less parallelism. The efficiency penalty of running LSTMs on a GPU can be mitigated by using GPU-LSTM implementations, such as the NVIDIA CUDA Deep Neural Network Library (cuDNN) to accelerate computations, and by using architectures with large output-input stride ratios so that the input strings to the LSTM layer are short.

[0131] In non-limiting embodiments, one or more models do not include an LSTM layer. For example, both p-CNN and ms-CNN variants do not include an LSTM layer. These models may have a finite ROI, and edge effects may only be possible within ROI / 2 from the window edge. In other words, multiple windows may overlap by approximately ROI / 2 input samples, and without using weighted windows, multiple windows may simply be concatenated after discarding half of each overlapping region. When applying such window usage, the efficiency benefits of longer windows may be more significant, especially considering the excellent parallelizability of CNN. In some examples, the longest possible window length may be used and a batch size of 1 may be applied to take system memory constraints into account. In some non-limiting embodiments, GPUs have achieved much greater guess rates than CPUs. However, when the model is small, i.e., there are only a few trainable parameters or the model is LSTM-based, CPU implementation may be preferred.

[0132] FIG. 18 illustrates the performance of one or more models according to non-limiting embodiments based on multiple sensors. In particular, the number of different sensor channels 1801 tested may include 15 accelerometers, 15 gyroscopes, 30 accelerometers and gyroscopes, 45 accelerometers, gyroscopes, and magnetometers, and 113 random sensor pairs. The models tested in FIG. 18 are deep convolutional LSTM 1802, p-CNN 1803, and ms-C / L 1804. As shown in FIG. 18, the model using accelerometers is more useful than gyroscopes, and the model using both accelerometers and gyroscopes also works well.

[0133] In a non-limiting embodiment, one or more models are fitted to a dataset with relatively few sensors. The models shown in FIG. 18 are trained and evaluated on the same training, validation, and test sets, but with different sets of sensor outputs ranging from 15 to 113 channels. The model architecture parameters are held constant or near constant, but the number of trainable parameters in the first model layer can vary as the number of input channels is changed. Further analysis can also be seen in FIG. 19, where the F for ms-C / L, ms-CNN, p-C / L, p-CNN, and b-LSTM are shown. 1w,nn and event-based F 1e Both are plotted for the same set of sensors.

[0134] The ms-C / L model outperforms other models, especially according to event-based metrics. The ms-C / L, ms-CNN, and p-C / L models perform well with fewer sensors. Five models have long or borderless ROIs, which may help the models compensate for missing sensor channels. In a non-limiting embodiment, one or more models work best with a 45 sensor set. This may indicate that the models may be overtrained for more than 45 sensor sets.

[0135] FIG. 19 illustrates a performance analysis of the model according to a non-limiting embodiment. In particular, FIG. 19 shows the per-sample F 1w,nn(1901) and various sensor combinations for event-based F 1e (1902), plotted for the same set of sensor tuples for ms-C / L, ms-CNN, p-C / L, p-CNN, and b-LSTM. Using a larger sensor tuples including all 113 standard sensor channels (All) plus the gyroscope (G), accelerometer (A), and magnetic (Mag) components of the Inertial Measurement Unit tended to improve performance metrics. Several models, including ms-C / L, ms-CNN, and p-C / L, maintained relatively high performance even with fewer sensor channels.

[0136] One or more models according to non-limiting embodiments can be used to simultaneously compute multiple independent outputs. For example, the same network can be used to predict both fast-changing behavior and slowly-changing poses simultaneously. The loss functions of the multiple outputs can simply be summed and the network can be trained on both simultaneously. This can allow some degree of automatic switching learning between the two label sets.

[0137] Non-limiting embodiments may be used to compute multi-label classification and regression problems by changing the output type, for example, changing the final activation function from softmax to sigmoid or linear and / or changing the loss function from cross-entropy to binary cross-entropy or mean squared error. In some examples, independent outputs of the same model may be combined. Also, in non-limiting embodiments, one or more other layers may be added. Other embodiments may help improve the layer modules by using skip connections or multi-field Inception-like architectures. Also, some non-limiting embodiments may be extended to real-time or streaming applications, for example, by using only CNNs or by replacing bidirectional LSTMs with unidirectional LSTMs.

[0138] While some of the data described above reflects pet activity data, in non-limiting embodiments, other data not reflecting pet activity may be processed and / or analyzed using the activity recognition time series classification algorithm to infer a desired output time series. For example, the other data may include, but is not limited to, financial data, cybersecurity data, electronic health records, audio data, image or video data, human activity data, or any other data known in the art. In such embodiments, the time series inputs may be from a wide range of different fields including finance, cybersecurity, electronic health records analysis, audio scene classification, and human activity recognition. The data may be, for example, time series data. Additionally or alternatively, the data may be data obtained from a first party, such as a wearable device, or third party data. Third party data may include data that is not collected directly by a particular company or entity and is purchased from another collecting entity or company. For example, third party data may be accessed or purchased using a data management platform. On the other hand, first party data may include data that is owned and / or collected directly by a particular company. For example, first party data may be collected from a consumer using a product or service provided by the particular company, such as a wearable device.

[0139] In one non-limiting embodiment, the time series classification algorithm may be applied to motor imagery electroencephalography (EEG) data. For example, EEG data may be collected while various subjects imagine performing one or more activities without physically performing them. Using the EEG observations, the time series classification algorithm may be trained to infer the activities the subjects are imagining. The determined classification may be used to create a brain-computer interface that allows a user to directly communicate with the outside world and / or control a tool using one or more imagined activities (also called brain intents).

[0140] The performance of the above examples may be demonstrated on various open source EEG intent recognition datasets, for example, the EEG movement imagery dataset from PhysioNet. See G. Schalk et al., “BCI2000: A General-Purpose Brain-Computer Interface (BCI) System,” IEEE Transactions on Biomedical Engineering, Issue 51(6), pages 1034-1043 (2004), available at https: / / www.physionet.org / content / eegmmidb / 1.0.0 / . In a non-limiting embodiment, no special spatial or frequency-based feature extraction methods were applied to the EEG movement imagery dataset. Instead, performance may be obtained by directly applying the model to 160Hz EEG observations. In some examples, the observations may be scaled to have zero mean and unit standard deviation according to the statistics of the training set. To ensure that the data is representative, data from each subject may be randomly split into a training set, a validation set, and a test set, and data from each subject may be represented in one set. Trials with subjects 1, 5, 6, 9, 14, 29, 32, 39, 42, 43, 57, 63, 64, 66, 71, 75, 82, 85, 88, 90, and 102 were used as test subjects, data from subjects 2, 8, 21, 25, 28, 40, 41, 45, 48, 49, 59, 62, 68, 69, 76, 81, and 105 were used for validation, and data from the remaining 70 subjects were used as training data. Each integer may represent one test subject. The performance of an example ms-C / L model is shown in Tables 1 and 2 below.

[0141] [Table 1]

[0142] [Table 2]

[0143] In a non-limiting embodiment, a system, method, or device may be used to assess the health of a pet. Data regarding the pet may be received as described above. The data may be received from at least one of the following data sources: wearable pet tracking or monitoring devices, genetic testing procedures, pet health records, pet insurance records, and / or input from the pet's owner. One or more of the above data sources may be collected using different sources. After the data is received, it may be compiled into one or more databases. This process or method may be performed by any device, hardware, software, algorithm, or cloud-based server described herein.

[0144] One or more health indicators of the pet may be determined based on the received data. For example, the health indicators may include metrics for licking, scratching, itching, walking, and / or sleeping by the pet. For example, the metrics may be the number of minutes per day the pet spends sleeping and / or the number of minutes per day the pet spends walking, running, or being otherwise active. Any other metric that may be indicative of the health of the pet may be determined. In some non-limiting embodiments, a health assessment of the pet may be performed based on the one or more health indicators. The health assessment may include, for example, evaluation and / or detection of skin conditions, skin diseases, ear / eye infections, arthritis, heart attacks, heart abnormalities, heart disease, allergies, dental abnormalities, dental disease, kidney abnormalities, kidney disease, cancer, endocrine abnormalities, endocrine diseases, hearing loss, depression, pancreatic attacks, pancreatic abnormalities, pancreatic diseases, obesity, metabolic abnormalities, metabolic diseases, and / or any combination thereof. The health evaluation may also include any other health abnormalities, diagnoses, or physical or mental diseases or disorders known in veterinary medicine.

[0145] Based on the health assessment, recommendations may be determined and transmitted to one or more of a pet owner, a veterinarian, a researcher, and / or any combination thereof. The recommendations may include, for example, one or more health recommendations to prevent the pet from developing one or more of a disease, an abnormality, a condition, and / or any combination thereof. The recommendations may include, for example, one or more of a food, a pet service, a dietary supplement, an ointment, a medicine to improve the health of the pet, a pet product, and / or any combination thereof. In other words, the recommendations may be nutritional recommendations. In some embodiments, the nutritional recommendations may include instructions to feed the pet with one or more of a chew, a dietary supplement, a food, and / or any combination thereof. In some embodiments, the recommendations may be medical recommendations. For example, the medical recommendations may include instructions to apply an ointment to the pet, administer one or more medicines to the pet, and / or provide one or more medicines to the pet.

[0146] The term "pet product" may include, for example, without limitation, any type of product, service, or device designed, manufactured, and / or intended for use by pets. For example, a pet product may be a toy, a chew, a food, a clothing, a collar, a medication, a health tracking device, a location tracking device, and / or any combination thereof. In another example, a pet product may include a genetic or DNA testing service for pets.

[0147] The term "pet owner" may include any person, organization, and / or group of people who owns and is responsible for any aspect of the care of a pet.

[0148] In non-limiting embodiments, a pet owner may purchase a pet insurance policy from a pet health care provider. To obtain the policy, the pet owner may pay a monthly, weekly, or yearly premium (also known as a premium). In some non-limiting embodiments, the premium and / or premium is determined in relation to a health assessment. In other words, the pet's health is determined, and the premium and / or premium that the policyholder (e.g., one or more pet owners) must pay may be determined based on the determined pet's health.

[0149] In other embodiments, premiums and / or discounts may be determined and applied to the base rate and / or premium of the pet health insurance policy. This determination may be automatic or manual. The premium and / or discount updates may be determined periodically, discretely, and / or continuously. For example, the premium or discount may be determined periodically every few months or weeks. In some non-limiting embodiments, the premium or discount may be determined based on data received after the recommendations are sent to one or more pet owners. In other words, the data may be used to monitor and / or track whether one or more pet owners are following and / or adhering to one or more provided recommendations. If the pet owner is following and / or adhering to one or more provided recommendations, a discount may be applied to the base rate or premium of the insurance policy. On the other hand, if one or more pet owners are not following and / or adhering to the provided recommendations, a premium and / or increase may be applied to the base rate or premium of the insurance policy. In non-limiting embodiments, the premium or discount to the base rate and / or premium may be determined based on one or more of the data, the health assessment, and / or the recommendations.

[0150] FIG. 20 illustrates a flow diagram of a process for assessing the health of a pet, according to a non-limiting embodiment. In particular, FIG. 20 illustrates a sequence of care that may include prediction 2001, prevention 2002, detection 2003, and treatment 2004. In prediction 2001, data may be used to understand or determine a particular pet's health condition or predisposition to a disease. This understanding or determination of the health condition or predisposition to a disease may be a health assessment. It is understood that the health assessment may be performed using any method described herein. The determined health condition or predisposition to a disease may be used to determine, calculate, or adjust the premium and / or the insurance policy. Prediction 2001 may be used to deliver or transmit a health assessment and / or a recommendation to the pet owner or any other interested party. For example, in some non-limiting embodiments, only a health assessment or a recommendation may be transmitted, while in other embodiments, both a health assessment and a recommendation may be transmitted. The recommendation may also be referred to as a health recommendation, a health alert, a health card, or a health report. In a non-limiting embodiment, a wearable device, such as a tracking or monitoring device, may be used to determine the recommendation.

[0151] Prevention 2002 as shown in FIG. 20 includes improving the premium margin of the pet insurance policy. This prevention step may help to improve the care of the pet and reward good pet owner behavior. In particular, prevention 2002 may provide recommendations to the pet owner that help the pet owner to better manage the health of the pet. The recommendations may be provided continuously, discretely, or periodically. After the recommendations are sent to the pet owner, data may be collected or received and used to track whether the pet owner is following the recommendations provided. This continuous monitoring after sending the recommendations may be compiled into a performance report. The performance report may be used to determine whether to adjust the base rate or premium of the pet insurance policy.

[0152] FIG. 20 also includes DETECTION 2003, which may be used to reduce intervention costs through early detection of possible pet health problems. As shown in PREVENTION 2002, recommendations may be sent or provided to the pet owner. These recommendations may help reduce intervention costs through early detection of possible health problems. In addition to or as part of the recommendations, telehealth services may be provided to the pet owner. The telehealth services may replace or be in addition to an in-person veterinary visit. The use of telehealth services may help reduce the costs and overhead associated with an in-person veterinary visit.

[0153] Treatment 2004 may include using the received or collected data to measure the effectiveness or value of early intervention for various diseases or health conditions. In a non-limiting embodiment, the data may detect a health indicator of the pet after the pet owner follows the recommendation. Based on the data, health indicator, or health assessment, the effectiveness of the recommendation may be determined. For example, the recommendation may include applying a tropical cream or ointment to the pet to treat the assessed skin condition. After applying the tropical cream or ointment, the collected data may help evaluate the effectiveness of treating the skin condition. In a non-limiting embodiment, a metric reflecting the effectiveness of the recommendation may be transmitted. The effectiveness of the recommendation may be, for example, clinical with respect to the pet and monetary with respect to the pet owner.

[0154] FIG. 21 illustrates example steps performed during a process for assessing a pet's health, according to a non-limiting embodiment. In particular, FIG. 21 illustrates an example of the prediction 2001 illustrated in FIG. 20. As previously described, the prediction 2001 can use data that can be used to understand or determine a particular pet's health abnormalities or predisposition to disease. The raw data can be used to calculate a health index, such as daily activity time or average scratching time. Based on that data, a health assessment of the pet can be determined. For example, as illustrated in FIG. 21, 37% of adult golden retrievers and Labrador retrievers with an average daily activity of 20-30 minutes are overweight or obese. Thus, if the health index indicates a low average daily activity, the corresponding health assessment can be that the pet is obese or overweight. The associated recommendation based on this health assessment can be to increase the average daily activity by at least 30-40 minutes.

[0155] A similar assessment can be made for scratching, which can be a health indicator, as shown in FIG. 21. If a particular pet is scratching beyond a threshold amount for a dog of a certain weight, breed, age, etc., the pet may have one or more of a skin disorder, a skin disease, another skin problem, and / or any combination thereof. Related recommendations can be provided to one or more of a pet owner, a veterinarian, a researcher, and / or any combination thereof. In some non-limiting embodiments, the health assessment can be used to determine the health of the pet and / or its predisposition to any health disorders and / or diseases. This health assessment can be used, for example, to determine the base or premium of an insurance policy.

[0156] FIG. 22 illustrates example steps performed during a process for evaluating a pet's health, according to non-limiting embodiments. In particular, FIG. 22 illustrates prevention 2002, shown in FIG. 20. Based on the health evaluation, a recommendation may be determined and sent to the pet owner. For example, as shown in FIG. 21, the recommendation may be to increase the pet's activity time. If the pet owner follows the recommendation and increases the pet's average daily activity level by the recommended amount, the premium or premium of the pet insurance policy may be lowered, reduced, or discounted. On the other hand, if the pet owner does not follow the recommendation, the premium or premium of the pet insurance policy may be increased or surcharged. In some non-limiting embodiments, additional recommendations or alerts may be sent to the user based on compliance or non-compliance with the recommendations. Recommendations or alerts may be personalized to the pet owner based on data collected about the particular pet. Recommendations or alerts may be provided to the pet owner periodically, such as daily, weekly, monthly, or yearly. As shown in FIG. 22, other health evaluations may include scratching or licking levels.

[0157] FIG. 23 illustrates example steps performed during a process for evaluating a pet's health, according to non-limiting embodiments. In particular, FIG. 23 illustrates the intervention 2003 shown in FIG. 20. As shown in FIG. 23, one or more health cards, reports, or alerts may be sent to the pet owner to communicate compliance or non-compliance with the recommendations. In some non-limiting embodiments, the pet owner may consult with a veterinarian or pet medical professional through a telehealth platform. Any telehealth platform known in the art may be used to facilitate communication between the pet owner and the veterinarian or pet medical professional. The telehealth visit may be included as part of the recommendations sent to the pet owner. The telehealth platform may operate on any user device, mobile device, or computer used by the pet owner.

[0158] Figure 24 illustrates example steps performed during a process for assessing a pet's health, according to a non-limiting embodiment. In particular, Figure 24 illustrates treatment 2004 shown in Figure 20. As shown in Figure 24, the data collected can be used to measure the economic and health benefits of the interventions recommended in Figures 21 and 22. For example, a comparison of health indicators such as scratching before and after a pet owner follows a recommendation can help assess the effectiveness of the recommendation.

[0159] FIG. 25A shows a flow diagram illustrating a process for assessing a pet's health, according to a non-limiting embodiment. In particular, FIG. 25A illustrates a method or process for data analysis performed using a system or device described herein. The method or process may include receiving data at a device, as shown at 2502. The data may include at least one of financial data, cybersecurity data, electronic health records, acoustic data, human activity data, and / or pet activity data. As shown at 2504, the method or process may include analyzing the data using two or more layer modules. Each layer module may include at least one of a many-to-many approach, striding, downsampling, pooling, multi-scaling, and batch normalization. In a non-limiting embodiment, each layer module may be represented by the following formula:

[0160]

number

[0161] Here, type is a convolutional neural network (CNN), and w out is the number of output channels, s is the stride ratio, k is the kernel length, and p drop is the dropout probability, b BN In some non-limiting embodiments, the two or more layers may include at least one of a full resolution convolutional neural network, a first pooling stack, a second pooling stack, a resampling step, a bottleneck layer, a regression stack, or an output module.

[0162] As shown at 2506, the method or process may include determining an output, such as a behavior classification or an intended action of the person, based on the analyzed data. The output may include a health assessment, a health recommendation, a financial prediction, or a security recommendation. At 2508, the method or process may include displaying the determined output on the mobile device.

[0163] FIG. 25B shows a flow diagram illustrating a process for assessing the health of a pet, according to a non-limiting embodiment. In particular, FIG. 25B illustrates a method or process performed using a system or device described herein. The method or process may include receiving data about the pet, as shown at 2512. The data may be received from at least one of a wearable pet tracking or monitoring device, a genetic testing procedure, a pet health record, a pet insurance record, or input from the pet owner. In some non-limiting embodiments, the data may reflect pet activity or behavior. The data may be received before and after a recommendation is sent to the pet owner's mobile device. Based on the data, one or more health indicators of the pet may be determined, as shown at 2514. The one or more health indicators may include metrics about licking, scratching, itching, walking, or sleeping by the pet.

[0164] A health assessment may be performed based on one or more health indicators of the pet, as shown at 2516. The one or more health indicators may be, for example, metrics about licking, scratching, itching, walking, or sleeping by the pet. At 2518, a recommendation may be sent to the pet owner based on the health assessment. The health assessment may include comparing the one or more health indicators to one or more stored health indicators. The stored health indicators are based on previous data about the pet and / or one or more other pets. For example, the recommendation may include one or more health recommendations to prevent the pet from developing, for example, one or more of an abnormality, disease, illness, and / or any combination thereof. In other non-limiting embodiments, the recommendation may include one or more of a food, a supplement, an ointment, a medicine, and / or any combination thereof that improves the health of the pet. In some non-limiting embodiments, the recommendation may include one or more of a recommendation to contact a telehealth service, a recommendation to request a telehealth visit, a notification for a telehealth appointment, a notification to set up a telehealth appointment, and / or any combination thereof. The recommendations may be sent to one or more mobile devices of one or more pet owners, veterinarians, and / or researchers, and displayed on the mobile devices of one or more pet owners, veterinarians, and / or researchers, as shown at 2520. The recommendations may be sent to the pet owners, veterinarians, and / or researchers on a periodic, discrete, or continuous basis.

[0165] In non-limiting embodiments, the effectiveness or effect of the recommendation may be determined or monitored based on the data. Metrics reflecting the effectiveness of the recommendation may be transmitted. The effect of the recommendation may be, for example, clinical with respect to the pet and financial with respect to the pet owner.

[0166] A premium or discount to be applied to the base rate or premium of the health insurance policy for the pet can be determined, as shown at 2522 in FIG. 25 . A discount to be applied to the base rate or premium of the health insurance policy can be determined when the pet owner complies with the recommendation. A premium to be applied to the base rate or premium of the health insurance policy can be determined when the pet owner does not comply with the recommendation. In non-limiting embodiments, the premium or discount to be applied to the base rate or premium of the health insurance policy can be provided to the pet owner or the health insurance policy provider. In some non-limiting embodiments, the base rate or premium of the health insurance policy for the pet can be determined based on a health evaluation. The determined premium or discount to be applied to the base rate or premium of the health insurance policy for the pet can be updated automatically or manually after the recommendation is sent to the pet owner, as shown at 2524.

[0167] In non-limiting embodiments, the health assessment and / or recommendation may be based on data including information about multiple pets. In other words, the health indicators of a particular pet may be compared to the health indicators of multiple other pets. Based on this comparison, a health assessment of the pet may be performed and appropriate recommendations may be provided. In some non-limiting embodiments, the health assessment and recommendation may be personalized based on the health indicators of a single pet. For example, instead of based on data collected from multiple other pets, the decision may be based on an algorithm or module that is tuned or trained based entirely or partially on data or information regarding the behavior of a single pet. Recommendations for pet products or services may be personalized to the behavior or specific health indicators of a single pet.

[0168] As described above, the health indicators may include, for example, metrics about licking, scratching, itching, walking, or sleeping by the pet. These health indicators may be determined based on data, information, or metrics collected from a wearable device having one or more sensors or accelerometers. The data collected from the wearable device may be processed by an activity recognition algorithm or model (also referred to as an activity recognition module or algorithm) to determine or identify the health indicators. The activity recognition algorithm or model may include two or more of the above layer modules. After the health indicators are identified, in non-limiting embodiments, the pet owner or caregiver may be asked to verify the correctness of the health indicators. For example, the pet owner or caregiver may receive a short message service, an alert or notification, such as a push alert, an email message on a mobile device, or any other type of message or notification. The message or notification may request the pet owner or caregiver to verify the health indicators identified by the activity recognition algorithm or model. In some non-limiting embodiments, the message or notification may indicate the time when the data, information, or metrics were collected. If the pet owner or caretaker is unable to verify the health indicators, the pet owner or caretaker may be asked to enter the pet's activity at the indicated time.

[0169] In a non-limiting embodiment, the pet owner or caretaker may be contacted after one or more health indicators are determined or identified. However, it is not necessary to contact the pet owner or caretaker after each health indicator is determined or identified. Contacting the pet owner or caretaker may be an automated process that does not require administrator intervention.

[0170] For example, the pet owner or caretaker may be contacted when the activity recognition algorithm or model has low confidence in the identified or determined health indicator, or when the identified health indicator is unusual, such as, for example, walking the pet at night or scratching for two consecutive hours. In some other non-limiting embodiments, the pet owner or caretaker need not be contacted when they are not around their pet at the time the health indicator is identified or determined. To determine that the pet owner or caretaker is not around their pet, the location reported from the pet owner's or caretaker's portable device may be compared to the location of the wearable device. Such a determination may utilize a short-range communication method, such as Bluetooth, or any other known method to determine the proximity of the portable device to the wearable device.

[0171] In some non-limiting embodiments, after one or more predefined health indicators are identified, the pet owner or caretaker may be contacted. The predefined health indicators may be selected, for example, based on a lack of training data or health indicators determined by the activity recognition algorithm or model with low precision or recall. The pet owner may then input a response, such as confirmation or rejection of the health indicator or activity, for example, using a GUI on the mobile device. The pet owner's or caretaker's response may be referred to as feedback. The GUI may list one or more pet activities or health indicators. The GUI may also include an option for the pet owner or caretaker to select from, neither rejecting nor confirming the health indicator or pet activity.

[0172] If the pet owner or caretaker confirms one or more health indicators during the indicated time, the activity training model may be further trained or adjusted based on the pet owner's or caretaker's confirmation. For example, input data used to train or adjust the activity recognition algorithm or model may be accelerometer or sensor data a predetermined time before or after the indicated time. The output of the activity recognition algorithm or model may be a high probability of about 1 that the pet licks over the indicated time period. The method, process, or system may record which activities the pet owner or caretaker did not confirm, and they may be ignored during the model training process.

[0173] In a non-limiting embodiment, the pet owner or caretaker can deny the occurrence of one or more health indicators at the indicated time and provide no information regarding the pet's activity at the indicated time. The pet owner can be the owner of the pet, and the caretaker can be any other person caring for the pet, such as a pet walker, veterinarian, or any other person watching over the pet. In such an embodiment, the input data used to train or tune the activity recognition algorithm or model can be accelerometer or sensor data a predetermined time before or after the indicated time. The output of the activity recognition algorithm or model can be a low probability of pet activity, about 0. The method, process, or system can record which activities the pet owner or caretaker denied, which can be ignored during the model training process.

[0174] In other non-limiting embodiments, the pet owner or caretaker can deny the occurrence of one or more health indicators at the indicated time and provide information regarding the pet's activity at the indicated time. In such embodiments, the input data used to train or tune the activity recognition algorithm or model can be accelerometer or sensor data before or after a predetermined time period before the indicated time. The output of the activity recognition algorithm or model can be a low probability of about 0 for the identified health indicator and a high probability of about 1 for the pet activity or health indicator entered by the pet owner or caretaker. The method, process, or system can record which activities the pet owner or caretaker denied, which can be ignored during the model training process.

[0175] In some non-limiting embodiments, the pet owner or caretaker does not deny or confirm the occurrence, in such embodiments, the pet owner or caretaker's responses or inputs may be excluded from the training set.

[0176] The responses or inputs provided by the pet owner or caretaker may be input into a training dataset of the activity recognition algorithm or model. The activity recognition module may be a deep neural network (DNN) trained using well-known DNN training techniques, such as stochastic gradient descent (SGD) or adaptive moment estimation (ADAM). In other embodiments, the activity recognition module may include one or more layer modules described herein. During training or tuning of the activity recognition model, health indicators not specified by the pet owner or caretaker may be removed from the model's calculation, while the associated classification loss may be appropriately weighted to help train the deep neural network. In other words, the deep neural network may be trained or tuned based on the pet owner's or caretaker's inputs. Training or tuning using the pet owner's or caretaker's inputs may help the deep neural network to better recognize the health indicators, thereby improving the accuracy of the health assessment and the associated recommendations. Training or tuning of the deep neural network based on the pet owner's or caretaker's responses may be based on sparse training, which allows the deep neural network to account for low quality data or partial data.

[0177] In non-limiting embodiments, the responses provided by the pet owner or caretaker are not used for simple correlation with the sensor or accelerometer data of the wearable device. The responses may be used to collect and annotate additional data that may be used to train the activity recognition model and improve the health assessment and / or recommendations provided. The incorporation of the pet owner or caretaker's responses into the training data may be automated. Such embodiments may be more efficient and less cost intensive than having to video confirm the determined or identified health indicators. In some non-limiting embodiments, the automated process may identify prediction failures of the activity recognition model, add the identified failures to a training database, and / or retrain or redeploy the activity recognition model. The prediction failures may be determined based on the responses provided by the pet owner or caretaker.

[0178] In some non-limiting embodiments, recommendations may be provided to the pet owner or caretaker based on the performed health assessment. The recommendations may include pet products or services. In non-limiting embodiments, the pet products or services may be determined and automatically sent to the pet owner or caretaker based on the recommendation. The pet owner or caretaker may apply or subscribe to automatic purchase and / or delivery of the recommended pet products or services. For example, the determined health indicator may be that the pet is scratching excessively based on data collected from the wearable device. Based on this health indicator, a health assessment may be performed and it may be determined that the pet has a skin problem. To address the skin problem, the recommendation may call for a skin and skin diet or a flea / tick removal product. The pet products associated with the recommendation may be automatically sent to the pet owner or caretaker without the pet owner or caretaker having to enter any information or approve the purchase or recommendation. In other non-limiting embodiments, the pet owner or caretaker may be asked to affirmatively approve the recommendation with an input. In addition to alerting the pet owner or caretaker, the health assessment and / or recommendation may be sent to a veterinarian. The transmissions to the veterinarian may also include recommendations to schedule a visit for the pet and recommended consultations via telemedicine services. In yet other embodiments, any other pet related content, instructions, and / or guidance may be transmitted to the pet owner, caretaker, or pet care provider, such as a veterinarian.

[0179] FIG. 26 shows a flow diagram illustrating a process for assessing a pet's health according to a non-limiting embodiment. In particular, FIG. 26 illustrates a method or process for data analysis performed using a system or device described herein. At 2602, an activity recognition model can be used to generate an event from the wearable device data. In other words, the activity recognition model can be used to determine or identify a health indicator based on data collected from the wearable device. At 2604, an event of interest, also referred to as a health indicator, can be identified. As shown at 2608, a pet owner or caretaker can be asked to confirm or deny whether a health indicator indicative of a pet's behavior or activity occurred at a indicated time. As shown at 2610 and 2612, based on the pet owner or caretaker's response, training examples can be generated and added to an updated training dataset. Then, as shown at 2614, the activity recognition model can be trained or adjusted based on the updated training dataset. Then, as shown at 2616, the trained or adjusted activity recognition model can be used to recognize one or more health indicators, perform a health assessment, and determine health recommendations. The trained or adjusted activity recognition model can be said to be customized or personalized to a particular pet.

[0180] As shown in FIG. 26 , determining the one or more health indicators may include processing the data with an activity recognition model. The one or more health indicators may be based on an output of the activity recognition model. The activity recognition model may be, for example, a deep neural network. The method or process may also include sending a request to a pet owner or caretaker to provide feedback on the one or more health indicators of the pet. The feedback is received from the pet owner or caretaker, and the activity recognition model may be trained or adjusted based on the feedback from the pet owner or caretaker. Additionally or alternatively, the activity recognition model may be trained or adjusted based on data from the one or more pets.

[0181] In non-limiting embodiments, the effectiveness of the recommendations may be determined. For example, after a recommendation is sent or displayed, the pet owner or caretaker may input or provide feedback indicating which of the one or more recommendations the pet owner or caretaker followed. In such embodiments, the pet owner or caretaker may indicate which recommendation they implemented and / or the date and / or time they began using the recommended product or service. For example, the pet owner or caretaker may begin feeding the recommended pet food to address a diagnosed or determined skin problem. The pet owner or caretaker may then indicate that they are using the recommended pet food and / or began using the food several days or weeks after the recommendation was sent or displayed on the mobile device. The feedback from the pet owner or caretaker may be used to track and / or determine the effectiveness of the recommendations. The effectiveness may then be reported to the pet owner or caretaker and / or additional recommendations may be made based on the determined effectiveness. For example, if the indicated pet food did not improve the tracked health indicators, a different pet product or service may be recommended. On the other hand, if the suggested pet food improves the tracked health indicator, it may be indicated to the pet owner or caretaker that the recommended pet food has improved the health of the pet.

[0182] As mentioned above, a tracking device according to the disclosed subject matter may consist of a computing device worn or otherwise carried by a user or other entity such as an animal. The wearable device may take any shape, form, color, or size. In one non-limiting embodiment, the wearable device may be attached to or placed on the pet in the form of a microchip. Additionally or alternatively, the tracking device may be a wearable device that can be coupled to a collar band, as described herein. For example, the wearable device may be attached to the pet's collar. FIG. 27 is a perspective view of a collar 2700 having a band 2710 with a tracking device 2720 attached thereto, according to an embodiment of the disclosed subject matter. The band 2710 includes a buckle 2740 and a clip 2730. FIG. 28 shows a perspective view of the tracking device 2720, and FIG. 29 shows a front view of the tracking device 2720.

[0183] As shown in FIG. 29, the wearable device 2720 can be rectangular in shape. In other embodiments, the wearable device 2720 can have any other suitable shape, such as an oval, square, or bone shape. The wearable device 2720 can have any suitable dimensions. For example, the device dimensions can be selected so that a pet can comfortably carry the device. For example, the wearable device weighs 0.91 ounces and is 1.4 inches (3.56 cm) wide, 1.8 inches (4.57 cm) high or long, and 0.6 inches (1.52 cm) thick or deep. In some non-limiting embodiments, the wearable device 2720 can be impact and / or water resistant.

[0184] The wearable device comprises a housing that may comprise a top cover 2721 and a base 2727 coupled to the top cover. The top cover 2721 has one or more sides 2723. As shown in the exploded view of FIGURE 30, the housing may further comprise internal mechanisms for the functional operation of the wearable device, such as a circuit board 2724 having a data tracking assembly and / or one or more sensors, a power source such as a battery 2725, a connector such as a USB connector 2726, internal hardware 2727 such as screws for coupling the device together, and other mechanisms.

[0185] The housing may further comprise an indicator, such as an illumination device (e.g., but not limited to, a light or light emitting diode), an acoustic device, and a vibration device. The indicator may be contained within the housing or disposed on the top cover of the device. As best shown in FIG. 29, an illumination device 2725 is depicted and embodied as a light on the top cover. However, the illumination device may be disposed within the housing and illuminate at least the top cover of the wearable device. In other embodiments, the acoustic device and / or the vibration device may comprise a tracking device. The acoustic device may include a speaker and may emit a sound, such as a whistle or speech, upon a trigger event. As described above, the indicator may be activated at a predefined geofence zone or boundary.

[0186] In a non-limiting embodiment, the lighting device 2725 may have different colors that indicate the battery charge level and / or the type of wireless access technology to which the wearable device 2720 is connected. In a non-limiting embodiment, the lighting device 2725 may be the lighting device described in FIG. 4. In other words, the lighting device 2725 may be manually or automatically activated when the pet leaves the geofence zone. Alternatively or additionally, the user may manually activate the lighting device 2725 using an application on the mobile device based on data received from the wearable device. Although the lighting device 2725 is shown as a light, in other embodiments not shown in FIG. 29, the lighting device may be replaced with a lighting device, a sound device, and / or a vibration device.

[0187] Figures 31-33 show side, top, and bottom views of a wearable tracking device 3000 similar to the wearable device 2720 shown in Figures 27-30. As depicted, the housing may further comprise an attachment 3002. The attachment 3002 may mate with a complementary receiving plate and / or collar band 2700 (described below with respect to Figure 42). The housing may further comprise a receiving port 3004 (described below with respect to Figure 33) for receiving a cable.

[0188] As shown in FIGS. 27-30, the top cover 2721 of the wearable device 2720 has a top surface and one or more side walls 2723 extending from the periphery of the top surface (best seen in FIGS. 28 and 30). In one non-limiting embodiment, the top cover is separable from the side walls and can be a separately manufactured and joined unit. In other embodiments, the top cover is integral with the side walls. The top cover can be made of a first material and the side walls can be made of a second material that is different from the first material. In other embodiments, the first and second materials are the same. In the embodiment of FIGS. 28 and 30, the top surface of the top cover 2721 is a different material than the one or more side walls 2723.

[0189] FIG. 34 depicts a perspective view of a tracking device according to another embodiment of the disclosed subject matter. In this embodiment, the top surface 3006 of the top cover is integral with and made of the same material as one or more side walls 3008. FIG. 35 shows a front view of the tracking device of FIG. 34. In this embodiment, the top cover includes an indicator embodied as a status identifier 3010. The status identifier may communicate the status of the device, such as a charging mode (reflecting a first color), a communicating mode (e.g., reflecting a second color when Bluetooth communication is running), and a fully charged mode (e.g., reflecting a third color when the battery life exceeds a threshold). For example, when the status identifier 3010 is amber, the wearable device may be charging. On the other hand, when the status identifier 3010 is green, it may be said that the battery of the wearable device is fully charged. In another example, the status identifier 3010 may be blue, meaning that the wearable device 3000 is connected via Bluetooth and / or is currently communicating with another device via a Bluetooth network. In a non-limiting embodiment, a wearable device using Bluetooth Low Energy (BLE) may be advantageous. BLE may be a wireless personal network that may help reduce power and resource consumption by the wearable device. Thus, using BLE may help extend the battery life of the wearable device. Other status modes of the status identifier 3010 and its colors are envisioned. The status identifier may also blink or have a selected pattern of blinking that may indicate a status. The top cover may include any suitable color and pattern and may further include a reflective or glow-in-the-dark material. FIG. 36 is an exploded view of the embodiment of FIG. 34, but with a different colored top cover for illustrative purposes. Similar to FIG. 30, the wearable device shown in FIG. 36 includes a circuit 3014, a battery 3016, a charging port 3018, a mechanical connector 3022, and / or a bottom cover 3020.

[0190] The housing, such as the top surface, may include indicia 3012, such as any suitable symbol, lettering, trademark, insignia, etc. As shown in the front view of Fig. 37, whistle indicia 3012 is shown on the top surface of the device. The housing may also include engravings representing personalized features, such as the wearer's name, or other identifying information, such as the pet owner's name and phone number. Figs. 38-40 show side, top, and bottom views of a tracking device 3000 that may further include such indicia if desired.

[0191] 41 illustrates a rear view of a tracking device to which a cable 4002 in accordance with the disclosed subject matter can be coupled. The cable 4002, such as a USB cable, can be inserted into the port 3004 to transmit data and / or charge the device.

[0192] As shown in Fig. 27, the tracking device may couple to a collar band 2700. The tracking device may couple to the band in any suitable manner known in the art. In one embodiment, the housing may couple to the collar band 2700 directly and / or via a complementary receiving plate 4004, for example, via an attachment on the base. Fig. 42 depicts an embodiment in which the band has a receiving plate 4004 that couples to the tracking device.

[0193] As shown in FIG. 27, the band 2700 may further include additional accessories known in the art. In particular, the band 2700 includes an adjustment mechanism 2730 for tightening or loosening the band, and may further include a clasp for coupling the band to a user, such as a pet. Any suitable fastening and adjustment structures are envisioned. FIGS. 43 and 44 depict an embodiment of the present tracking device coupled to a pet P, such as a dog, via a collar band. As shown in FIG. 44, the band may further include additional accessories, such as a name tag 4460.

[0194] As discussed above, FIG. 45 depicts a receiving plate and / or support frame and collar band 2700. The support frame can be used to couple a tracking device to the collar band 2700. Attachments for use with tracking devices in accordance with the disclosed subject matter are described in U.S. Provisional Patent Application No. 62 / 768,414, filed Nov. 16, 2018, the entire contents of which are incorporated herein by reference. As embodied herein, the support frame can include receiving holes and latches for coupling with the attachments and / or inserts of the tracking device.

[0195] The collar band 2700 may be coupled to a support frame. By way of example, as embodied in FIG. 42, the collar band may include loops for coupling to the support frame. Additionally or alternatively, it may be desirable to couple a tracking device according to the disclosed subject matter to a collar without loops or other suitable structure for securing the support frame. With reference to FIGS. 45-47, the support frame may include a collar attachment feature for attaching the support frame to a collar.

[0196] For example, referring to FIG. 45, the support frame may include a hook and loop collar attachment feature. A strap 4502 may be attached to a bar 4506 of the support frame. The strap 4502 may include a hook portion 4504 having a number of hooks and a loop portion 4503 having a number of loops. The support frame 4501 may be secured to a collar (not shown) by wrapping the strap 4502 around the collar and then wrapping the strap around the bar 4505 of the support frame 4501. After the strap 4502 is wrapped around the collar and bar 4505, the hook portion 4504 may engage with the loop portion 4503 to secure the support frame 4501 to the collar. Although the strap 4502 has been described as being used to secure the support frame 4501 to a collar, it should be understood that the strap 4502 may also function as a collar. The length of the strap 4502 may be adjusted based on the desired configuration of the attachment features.

[0197] 46, the support frame 4601 may be secured to the collar using snap members 4602. As embodied herein, the support frame 4601 may have grooves 4603 configured to receive tabs 4604 on the snap members 4602. The support frame may be secured to the collar (not shown) by passing the collar through channels 4605 in the snap members 4602 and engaging the tabs of the snap members with the grooves 4603 in the support frame 4601. The tabs 4604 may have a lip or ridge that prevents the snap members 4612 from separating from the support frame 4601.

[0198] Additionally or alternatively, with reference to FIG. 47, the support frame 4701 may be secured to the collar using a strap 4703 having a bar 4702. As embodied herein, the support frame 4701 may have channels 4704 on either side thereof. The channels 4704 may be configured to receive and hold the bar 4702. The bar 4702 may be attached to the strap 4703. By way of example, the strap 4703 may be made of a flexible material such as rubber. The support frame 4701 may be secured to the collar by wrapping the strap 4703 around the collar (not shown) and attaching the bar 4702 within the channel 4704 of the support frame 4701.

[0199] For purposes of disclosure, a module is a hardware, software, or firmware (or combination thereof) system, process or function, or part thereof, that performs or enables (with or without human intervention or assistance) the processes, features, and / or functions described herein. A module may include sub-modules. Software parts of a module may be stored on a computer-readable medium for execution by a processor. A module may be integrated into one or more servers or deployed and executed by one or more servers. One or more modules may be organized into an engine or application.

[0200] For purposes of disclosure, the terms "user," "subscriber," "consumer," or "customer" should be understood to refer to a user of an application as described herein and / or a consumer of data provided by a data provider. By way of example and not limitation, the terms "user" or "subscriber" may refer to a person who receives data provided by a data or service provider when browsing over the Internet, or may refer to an automated software application that receives, stores, or processes the data.

[0201] Those skilled in the art will appreciate that the methods and systems of the present disclosure may be implemented in many ways and should not be limited by the exemplary embodiments and examples described above. In other words, functional elements may be performed by one or many pieces of hardware and software or firmware in various combinations, and individual functions may be distributed among multiple software applications at the client level or server level or both. In this regard, any number of features of the different embodiments described herein may be combined into one or many embodiments, and other embodiments having less than or more than all of the features described herein are possible.

[0202] Functions may also be distributed, in whole or in part, among multiple components in ways that are or will become known. Thus, a myriad of software / hardware / firmware combinations are possible to implement the functions, features, connections, and preferences described herein. Those skilled in the art will also recognize that the scope of the present disclosure includes conventionally known ways of implementing the described features and functions and connections, as well as modifications and variations that may be made to the hardware or software or firmware components described herein.

[0203] Additionally, the method embodiments presented and described in the flowcharts of the present disclosure are provided as examples to provide a more complete understanding of the present technology. The method of the present disclosure is not limited to the operations and logic flows presented herein. Other embodiments are envisioned in which the order of various operations is changed and sub-operations that are part of a larger operation are performed independently.

[0204] While various embodiments have been described for purposes of disclosure, such embodiments should not be construed as limiting the teachings of the disclosure to those embodiments. Various modifications and variations may be made to the elements and operations described above with results that remain within the scope of the systems and processes described in this disclosure.

[0205] Although the disclosed subject matter has been described using preferred embodiments, those skilled in the art will understand that various modifications and improvements can be made to the disclosed subject matter without departing from its scope. Additional features known in the art, such as those disclosed in U.S. Pat. No. 10,142,773, U.S. Patent Application Publication No. 2014 / 0290013, U.S. Design Application Nos. 29 / 670,543, 29 / 580,756, and U.S. Provisional Patent Application Nos. 62 / 768,414, 62 / 867,226, and 62 / 970,575, may also be incorporated. Each document is incorporated herein by reference in its entirety. Also, although individual features of one non-limiting embodiment of the disclosed subject matter may be described herein and shown in the drawings of that non-limiting embodiment rather than in other embodiments, it should be apparent that individual features of one non-limiting embodiment may be combined with one or more features of another embodiment or with features of multiple embodiments.

[0206] Preferred embodiments of the present invention will be described below in detail.

[0207] EMBODIMENT 1 1. A computer-implemented method for monitoring pet activity, comprising: receiving data about the pet from a wearable device comprising a sensor; determining one or more health indicators of the pet based on the data; and performing a health assessment of the pet based on the one or more health indicators of the pet; The method includes: EMBODIMENT 2 2. The method of embodiment 1, further comprising the step of transmitting the health assessment of the pet to a mobile device. EMBODIMENT 3 3. The method of embodiment 2, further comprising displaying the health assessment of the pet using a graphical user interface on the portable device. EMBODIMENT 4 The step of determining one or more health indicators of the pet based on the data includes: processing the data with an activity recognition model; determining the one or more health indicators based on an output of the activity recognition model; 4. The method according to any one of embodiments 1 to 3, further comprising: EMBODIMENT 5 5. The method of embodiment 4, wherein the activity recognition model is a deep neural network. EMBODIMENT 6 6. The method of embodiment 5, wherein the deep neural network comprises two or more layer modules, each of the layer modules including at least one of a many-to-many approach, striding, downsampling, pooling, multi-scaling, or batch normalization. EMBODIMENT 7 Each layer module is represented by the following notation, where type is a convolutional neural network (CNN), and w out is the number of output channels, s is the stride ratio, k is the kernel length, and p drop is the dropout probability, and b BN 7. The method of embodiment 6, wherein:

[0208]

number

[0209] EMBODIMENT 8 sending a request to a pet owner or caretaker to provide feedback on the one or more health indicators of the pet; receiving said feedback from said pet owner or caretaker; 8. The method of any one of embodiments 1 to 7, further comprising: EMBODIMENT 9 9. The method of embodiment 8, further comprising training or adjusting the pet activity recognition model based on the feedback from the pet owner or caretaker. EMBODIMENT 10 10. The method according to any of embodiments 4-9, further comprising training or tuning the pet activity recognition model based on data from one or more other pets. EMBODIMENT 11 The method according to any of the preceding claims, wherein the method is performed by at least one of the wearable device, one or more servers, or a cloud computing platform. EMBODIMENT 12 12. The method of any of the preceding claims, wherein the one or more health indicators comprise metrics about licking, scratching, itching, walking, or sleeping by the pet. EMBODIMENT 13 13. The method according to any of the preceding claims, wherein the data includes the location of the pet, the location being determined using a global positioning system. EMBODIMENT 14 14. The method according to any of the preceding claims, further comprising instructing the wearable device to turn on a lighting device based on the health assessment of the pet. EMBODIMENT 15 The step of performing the health assessment A method as described in any one of embodiments 1 to 14, comprising the step of comparing the health index to a stored health index, the stored health index being based on previous data relating to the pet or one or more other pets. EMBODIMENT 16 16. The method of any one of the preceding claims, wherein the sensor comprises at least one of an actuator, a gyroscope, a magnetometer, a microphone, or a pressure sensor. EMBODIMENT 17 determining a health recommendation or fitness prompt for the pet based on the health assessment; transmitting the health recommendation or fitness prompt to the mobile device; 17. The method of any one of the preceding embodiments, further comprising: EMBODIMENT 18 18. The method of any one of the preceding embodiments, wherein the health recommendations include pet food or pet product recommendations. EMBODIMENT 19 19. The method according to any of the preceding claims, wherein the data relating to the pet is received continuously or discretely from the wearable device. EMBODIMENT 20 20. The method according to any of the preceding embodiments, wherein the wearable device tracks the data about the pet live. EMBODIMENT 21 1. A wearable device, comprising: The top cover and a base connected to the top cover; a housing having the housing includes a sensor for monitoring data related to a pet, a transceiver for transmitting the data related to the pet, and a display; The wearable device, wherein the indicator is at least one of a lighting device, a sound device, or a vibration device. EMBODIMENT 22 22. The wearable device of embodiment 21, wherein the indicator provided within the housing is configured to be turned on after the wearable device leaves a geofence zone. EMBODIMENT 23 A wearable device as described in embodiment 21 or 22, wherein the data regarding the pet is used for at least one of the steps of determining one or more health indicators of the pet or performing a health assessment of the pet. EMBODIMENT 24 A wearable device as described in embodiment 23, wherein the wearable device is configured to transmit the data regarding the pet to one or more servers or cloud computing platforms, and the step of determining the one or more health indicators is performed by at least one of the wearable device, the one or more servers, or the cloud computing platform. EMBODIMENT 25 A wearable device as described in any of embodiments 21 to 24, wherein the data regarding the pet is transmitted to one or more servers or cloud computing platforms. EMBODIMENT 26 A wearable device according to any one of embodiments 21 to 25, wherein the display is disposed on the top cover. EMBODIMENT 27 27. The wearable device of any one of embodiments 21 to 26, wherein the lighting device includes an electric lamp or a light-emitting diode. EMBODIMENT 28 A wearable device described in any of embodiments 21 to 27, wherein the lighting device is disposed within the housing and configured to illuminate at least the top cover of the wearable device. EMBODIMENT 29 The wearable device of any one of embodiments 21 to 28, wherein the top cover has an upper surface and a side wall extending from an outer periphery of the upper surface. EMBODIMENT 30 30. The wearable device of any one of embodiments 21 to 29, wherein the top cover is integral with the side wall. EMBODIMENT 31 A wearable device described in any of embodiments 21 to 30, wherein the housing has a receiving port for receiving a cable. EMBODIMENT 32 A wearable device as described in any one of embodiments 21 to 31, wherein the housing is provided with an attachment that is coupled to a collar band.

[0210] EMBODIMENT 33 1. A method for monitoring pet activity, comprising: monitoring a position of the wearable device; determining that the wearable device has left a geofence zone based on a position of the wearable device; instructing the wearable device to turn on a display after determining that the wearable device has left the geofence zone; wherein the indicator is at least one of a lighting device, a sound device, or a vibration device.

[0211] EMBODIMENT 34 determining that the wearable device has entered the geofence zone; turning off the indicator after the wearable device enters the geofence zone. 34. The method of embodiment 33, further comprising:

[0212] EMBODIMENT 35 35. The method of embodiment 34, wherein the determination that the wearable device has entered the geofence zone is a determination that the wearable device has exited the geofence zone and then re-entered the geofence zone. EMBODIMENT 36 receiving an instruction from a portable device to turn off the indicator; 36. The method according to any one of embodiments 33 to 35, wherein the mobile device comprises an application that allows a user to turn off the display. EMBODIMENT 37 A method as described in any one of embodiments 33 to 36, wherein the step of instructing the wearable device to turn on the display includes turning on a light or light-emitting diode of the lighting device. EMBODIMENT 38 A method according to any one of embodiments 33 to 37, further comprising receiving the position of the wearable device by a Global Positioning System (GPS) receiver. EMBODIMENT 39 The step of monitoring a position of the wearable device A method according to any one of embodiments 33 to 38, further comprising identifying active wireless networks within the vicinity of the wearable device. EMBODIMENT 40 40. The method of embodiment 39, wherein the step of determining that the wearable device has left the geofence zone may include determining that the active wireless network is no longer within the vicinity of the wearable device. EMBODIMENT 41 41. The method of any one of embodiments 33 to 40, wherein the geofence zone is predetermined using latitude and longitude coordinates.

[0213] EMBODIMENT 42 1. A method for data analysis comprising: receiving data at a device; analyzing the data using two or more layer modules, each layer module including at least one of a many-to-many approach, striding, downsampling, pooling, multi-scaling, or batch normalization; determining an output, such as a behavior classification or an intended behavior of a person, based on the analyzed data; The method includes: EMBODIMENT 43 43. The method of embodiment 42, wherein the data includes at least one of financial data, cybersecurity data, electronic health records, image or video data, audio data, human activity data, or pet activity data. EMBODIMENT 44 44. The method of embodiment 42 or 43, wherein the output comprises a health assessment, health recommendations, financial predictions, image or video recognition, voice recognition, or security recommendations. EMBODIMENT 45 Each layer module is represented by the following notation, where type is a convolutional neural network (CNN), and w out is the number of output channels, s is the stride ratio, k is the kernel length, and p drop is the dropout probability, and b BN The method according to any one of embodiments 42 to 44, wherein:

[0214]

number

[0215] EMBODIMENT 46 46. ​​The method of any one of embodiments 42 to 45, wherein the two or more layers include at least one of a full resolution convolutional neural network, a first pooling stack, a second pooling stack, a resampling step, a bottleneck layer, a regression stack, or an output module. EMBODIMENT 47 47. The method of any one of embodiments 42 to 46, wherein the data is time series data. EMBODIMENT 48 48. The method according to any one of embodiments 42-47, further comprising the step of displaying the determined output on a mobile device. EMBODIMENT 49 1. A computer-implemented method for assessing the health of a pet, comprising: receiving data about the pet; determining one or more health indicators of the pet based on the data; and performing a health assessment of the pet based on the one or more health indicators; The method includes: EMBODIMENT 50 The method of embodiment 48, further comprising determining a recommendation for the pet owner based on said health assessment. EMBODIMENT 51 51. The method of embodiment 50, further comprising the step of transmitting the recommendation to a mobile device of the pet owner. EMBODIMENT 52 52. The method of embodiment 51, further comprising the step of displaying the recommendation to a user on the mobile device using a graphical user interface. EMBODIMENT 53 53. The method of any of embodiments 49-52, further comprising determining a premium or discount to be applied to the base or premium of the pet health insurance policy based on the data. EMBODIMENT 54 54. The method of any of embodiments 49-53, further comprising the step of providing the pet owner or the provider of the health insurance policy with a premium or discount applied to the base rate or premium of the health insurance policy. EMBODIMENT 55 A method according to any of embodiments 49 to 54, wherein the data is received before and after the recommendation is sent to the pet owner's mobile device. EMBODIMENT 56 56. The method according to any of embodiments 49-55, further comprising determining a base or premium for the health insurance policy for the pet based on the health assessment. EMBODIMENT 57 A method as described in any of embodiments 49 to 56, further comprising a step of automatically or manually updating the determined surcharge or discount applied to the basic fee or premium of the health insurance policy for the pet based on the data received after the recommendation is sent to the pet owner. EMBODIMENT 58 a discount to be applied to the base rate or premium of the health insurance policy is determined if the pet owner complies with the recommendation; The method according to any one of embodiments 49 to 57, wherein a surcharge to be applied to the basic fee or premium of the health insurance policy is determined if the pet owner does not comply with the recommendation. EMBODIMENT 59 The method of any of embodiments 49 to 58, further comprising determining a premium or discount on the base rate or premium of the health insurance policy for the pet based on at least one of the data, the health assessment, and the recommendation. EMBODIMENT 60 The method of any one of embodiments 49-59, wherein the one or more health indicators comprise metrics for licking, scratching, itching, walking, or sleeping by the pet. EMBODIMENT 61 The method of any of embodiments 49-60, wherein the recommendations include one or more health recommendations to prevent the pet from developing a disease or ailment. EMBODIMENT 62 The method of any of embodiments 49 to 61, further comprising receiving the data from at least one of a wearable pet tracking or monitoring device, a genetic testing procedure, a pet health record, a pet insurance record, or input from the pet owner. EMBODIMENT 63 The step of performing the health assessment A method as described in any of embodiments 49 to 62, comprising a step of comparing the health index to a stored health index, the stored health index being based on previous data regarding the pet or one or more other pets. EMBODIMENT 64 64. The method according to any one of embodiments 49 to 63, wherein the recommendations are sent to the pet owner periodically or continuously. EMBODIMENT 65 determining or monitoring the validity of said recommendation based on said data; transmitting a metric reflecting the validity of the recommendation; 65. The method of any one of embodiments 49-64, further comprising: wherein the benefit of the recommendation is clinical with respect to the pet or financial with respect to the pet owner. EMBODIMENT 66 The method of any one of embodiments 49-65, wherein the recommendation comprises a food, a dietary supplement, an ointment, or a medicine that will improve the health of the pet. EMBODIMENT 67 67. The method of any one of embodiments 49-66, wherein the recommendation includes a remote health service or visit. EMBODIMENT 68 The step of determining one or more health indicators of the pet based on the data includes: processing the data with an activity recognition model; determining the one or more health indicators based on an output of the activity recognition model; 68. The method of any one of embodiments 49-67, further comprising: EMBODIMENT 69 69. The method of embodiment 68, wherein the activity recognition model is a deep neural network. EMBODIMENT 70 70. The method of embodiment 69, wherein the deep neural network comprises two or more layer modules, each of which includes at least one of a many-to-many approach, striding, downsampling, pooling, multi-scaling, or batch normalization. EMBODIMENT 71 Each layer module is represented by the following notation, where type is a convolutional neural network (CNN), and w out is the number of output channels, s is the stride ratio, k is the kernel length, and p drop is the dropout probability, and b BN 71. The method of embodiment 70, wherein:

[0216]

number

[0217] EMBODIMENT 72 An apparatus comprising: at least one memory having computer program code; At least one processor and Equipped with An apparatus, wherein the at least one memory having the computer program code is configured to cause the apparatus to execute, using the at least one processor, a method according to any one of embodiments 1 to 20 and 32 to 71. EMBODIMENT 73 A non-transitory computer-readable medium storing instructions which, when executed by hardware, perform a process according to the method of any one of embodiments 1 to 20 and 32 to 71. EMBODIMENT 74 A computer program product encoded with instructions for carrying out a process according to the method of any one of embodiments 1 to 20 and 32 to 71. [Explanation of symbols]

[0218] 100 Systems 102 Tracking Device 104 Portable Devices 106 Server 108 Network 200, 300 devices 204 Memory 206 Non-volatile storage devices 208 Sensors 210 GPS receiver 212 Cellular Transceiver 214 Radio Transmitter / Receiver 216 Bluetooth Transmitter / Receiver 302 GPS receiver 304 Geofence Detector 306 Sensor 308 Storage device 312 Network Interface 601 Many to One Model 602 Many-to-Many Model 701 Shedding layer 702 1D Average Pooling 703 1D Bulk Normalization 711 Full Resolution CNN 712 1st Pooling Stack 713 Second Pooling Stack 714 resampling steps 715 Bottleneck Layer 716 Regression Stack 717 Output Module 721 Fine-scale CNN 722 Coarse-scale RNN stack 723 Mixed Scale Final Stack

Claims

1. 1. A wearable device, comprising: The top cover and a base connected to the top cover; a housing having the housing includes a sensor for monitoring data related to a pet, a transceiver for transmitting the data related to the pet, and a display; the indicator is at least one of a lighting device, a sound device, or a vibration device; The wearable device, wherein the indicator provided within the housing is configured to be turned on after the wearable device leaves a geofence zone.

2. 2. The wearable device of claim 1, wherein the data regarding the pet is used for at least one of the steps of determining one or more health indicators of the pet including metrics about licking, scratching, itching, walking, or sleeping by the pet, comprising processing the data through an activity recognition model including an LSTM model, a CNN model, or a combination thereof, and determining the one or more health indicators based on an output of the activity recognition model; and performing a health assessment of the pet based on the one or more health indicators of the pet.

3. The wearable device of claim 2, wherein the wearable device is configured to transmit the data regarding the pet to one or more servers or cloud computing platforms, and the step of determining the one or more health indicators is performed by at least one of the wearable device, the one or more servers, or the cloud computing platform.

4. The following conditions: (i) the data about the pet is transmitted to one or more servers or cloud computing platforms; (ii) the indicator is disposed on the top cover; (iii) the lighting device comprises an electric lamp or a light emitting diode; (iv) the illumination device is disposed within the housing and configured to illuminate at least the top cover of the wearable device; (v) the top cover has a top surface and a sidewall extending from an outer periphery of the top surface; (vi) the top cover is integral with the side wall; (vii) the housing includes a receiving port for receiving a cable; and (viii) the housing includes an attachment adapted to be coupled to a collar band; The wearable device according to any one of claims 1 to 3, wherein at least one of the following is satisfied.

5. 1. A method for monitoring pet activity, comprising: monitoring a position of the wearable device; determining that the wearable device has left a geofence zone based on a position of the wearable device; instructing the wearable device to turn on an indicator after determining that the wearable device has left the geofence zone; wherein the indicator is at least one of a lighting device, a sound device, or a vibration device.

6. determining that the wearable device has entered the geofence zone; turning off the indicator after the wearable device enters the geofence zone. The method of claim 5 further comprising:

7. 7. The method of claim 6, wherein the determination that the wearable device has entered the geofenced zone is a determination that the wearable device has exited the geofenced zone and then re-entered the geofenced zone.

8. receiving an instruction from a portable device to turn off the indicator; The method according to any of claims 5 to 7, wherein the mobile device comprises an application allowing a user to turn off the display.

9. The method according to any of claims 5 to 8, further comprising the step of receiving the position of the wearable device by a Global Positioning System (GPS) receiver.

10. The following conditions: (i) the step of instructing the wearable device to turn on the indicator comprises turning on a light or light emitting diode of the illumination device; (ii) the step of monitoring the position of the wearable device identifying active wireless networks within a vicinity of the wearable device. (iii) the step of determining that the wearable device has left the geofence zone may include determining that the active wireless network is no longer within a vicinity of the wearable device; and (iv) the geofence zone is predetermined using latitude and longitude coordinates; The method according to any one of claims 5 to 9, wherein at least one of the following is satisfied.

11. 1. A method for data analysis comprising: receiving data at a device; analyzing the data using two or more layer modules, each layer module including at least one of a many-to-many approach, striding, downsampling, pooling, multi-scaling, or batch normalization; determining an output, such as a behavior classification or an intended behavior of a person, based on the analyzed data; The method includes:

12. The following conditions: (i) the data includes at least one of financial data, cybersecurity data, electronic health records, image or video data, audio data, human activity data, or pet activity data; (ii) the output comprises a health assessment, a health recommendation, a financial prediction, an image or video recognition, a voice recognition, or a security recommendation; (iii) Each layer module is represented by the following notation, where type is a convolutional neural network (CNN), and w out is the number of output channels, s is the stride ratio, k is the kernel length, and p drop is the dropout probability, and b BN is the collective normalization, [0025] (iv) the two or more layers include at least one of a full resolution convolutional neural network, a first pooling stack, a second pooling stack, a resampling step, a bottleneck layer, a regression stack, or an output module; and (v) the data is time series data; The method of claim 11 , wherein at least one of the following is satisfied:

13. 13. The method of claim 11 or 12, further comprising the step of displaying the determined output on a portable device.

14. 1. A computer-implemented method for assessing the health of a pet, comprising: receiving data about the pet; determining one or more health indicators of the pet based on the data, including metrics on licking, scratching, itching, walking, or sleeping by the pet, comprising processing the data through an activity recognition model, including an LSTM model, a CNN model, or a combination thereof, and determining the one or more health indicators based on an output of the activity recognition model; performing a health assessment of the pet based on the one or more health indicators; Including, (i) determining a premium or discount to be applied to the base rate or premium of the pet health insurance policy based on the data; (ii) offering the pet owner or the health insurance policy provider a premium or discount applicable to the base rate or premium of the health insurance policy; (iii) determining a base rate or premium for the health insurance policy for the pet based on the health evaluation; (iv) automatically or manually updating the determined premium or discount applied to the base rate or premium of the health insurance policy for the pet based on the data received after the recommendation is sent to the pet owner; and (v) determining a premium or discount for the base rate or premium of the health insurance policy for the pet based on at least one of the data, the health assessment, and the recommendation; The method further comprises:

15. a discount to be applied to the base rate or premium of the health insurance policy is determined if the pet owner complies with the recommendation; 15. The method of claim 14, wherein a premium or surcharge to be applied to the health insurance policy is determined if the pet owner does not comply with the recommendation.

16. determining or monitoring the validity of said recommendation based on said data; transmitting a metric reflecting the validity of the recommendation; 16. The method of claim 14 or 15, further comprising: wherein the benefit of the recommendation is clinical with respect to the pet or financial with respect to the pet owner.

Images