Entity identification methods, computer program products, and entity identification devices implemented by computers or hardware.

Abstract

The present disclosure relates to a computer- or hardware-implemented method (100) for identifying entities, comprising the steps of: a) providing (110) inputs from a plurality of sensors to a network of a plurality of nodes; b) generating (120) an activity level by each node of the network based on the inputs from the plurality of sensors; c) comparing (130) the activity level of each node to a threshold level; d) setting (140) the activity level for each node to a preset value or retaining the generated activity level based on the comparing step; e) calculating (150) a total activity level as the sum of all activity levels of the nodes of the network; f) repeating (160) steps a)-e) until a local minimum in the total activity level is reached; and g) once a local minimum in the total activity level is reached, utilizing (170) the distribution of activity levels at the local minimum to identify measurable entity characteristics. The present disclosure further relates to a computer program product and an entity identification device (300).

Description

Technical Field

[0001] The present disclosure relates to an entity identification method implemented by a computer or hardware, a computer program product, and an entity identification device. More specifically, the present disclosure relates to an entity identification method implemented by a computer or hardware, a computer program product, and an entity identification device as described in the superordinate concepts of claims 1, claim 9, and claim 10.

[0002] Background Art Entity identification is known from the prior art. One technique used to perform entity identification is a neural network. One type of neural network available for entity identification is a Hopfield neural network. The Hopfield neural network is a form of recurrent artificial neural network. The Hopfield neural network functions as a memory system that has binary threshold nodes and is addressable by content ( "associative").

[0003] However, the performance of existing neural network-based solutions is insufficient and / or the reliability is low. Furthermore, existing solutions require a great deal of time for training, and thus may require a large amount of computer power and / or energy, especially for training. Moreover, existing neural network-based solutions may require a large amount of storage space.

[0004] [[ID=I9]]Therefore, an alternative approach to entity identification is needed. Preferably, such an approach results in or enables one or more of improved performance, even higher reliability, increased efficiency, even faster training, even less use of computer power, even less use of training data, even less use of storage space, and / or even less use of energy.

[0005] overview One objective of this disclosure is to mitigate, alleviate or eliminate one or more of the aforementioned shortcomings and inconveniences of the prior art, and to resolve at least the aforementioned problems. According to a first aspect, an entity identification method is provided that is implemented by computer or hardware, the method comprising the steps of: a) supplying inputs from multiple sensors to a network of multiple nodes; b) having each node of the network generate an activity level based on the inputs from the multiple sensors; c) comparing the activity level of each node with a threshold level; d) setting the activity level of each node to a preset value or retaining the generated activity level based on the comparison step; e) calculating a total activity level as the sum of all activity levels of the nodes in the network; f) repeating steps a) to e) until a local minimum of the total activity level is reached; and g) once a local minimum of the total activity level is reached, utilizing the distribution of activity levels at the local minimum to identify measurable entity characteristics. The advantage of the first embodiment is that the effective structure of the network can be dynamically changed, thereby enabling the identification of a greater number of entities, for example, per unit / node.

[0006] In some embodiments, the input changes dynamically over time and follows a single sensor input trajectory. One advantage of these embodiments is that this method is less susceptible to noise. Another advantage is that identification is faster. Yet another advantage is that this enables more accurate identification.

[0007] In some embodiments, multiple sensors monitor the dependencies between them. The advantage of this method is that it is less susceptible to noise.

[0008] According to some embodiments, the state in which the total activity level reaches a local minimum is reached when a single sensor input trajectory is followed for a period longer than a user-definable time threshold and with a deviation smaller than a user-definable deviation threshold.

[0009] According to some embodiments, the activity level of each node is used for all other nodes as a weighted input, each weighted by the other, where at least one weighted input is negative and / or at least one weighted input is positive and / or all held generating activity levels are positive scalars.

[0010] According to some embodiments, the network is activated by an activation energy X that influences where the local minimum of the total activity level lies.

[0011] According to some embodiments, inputs from multiple sensors are pixel values, e.g., luminance, of images captured by a camera, and the distribution of activity levels across all nodes is further used to control the camera's position by rotational and / or translational motion, thereby controlling the sensor input trajectory, and the entities identified are objects or features of objects present in at least one of the captured images. One advantage of these embodiments is that this method is less susceptible to noise. Another advantage is that identification does not depend on absolute time, for example, the absolute time spent by each still camera image, and the time spent between various such still camera images as the camera position changes.

[0012] According to some embodiments, the multiple sensors are touch sensors, and the input from each of the multiple sensors is a touch event signal having a force-dependent value, and the distribution of activity levels across all nodes is used to identify the sensor input trajectory as a new contact event, the end of a contact event, a gesture, or applied pressure.

[0013] According to some embodiments, each sensor of a plurality of sensors is associated with a different frequency band of a single audio signal, each sensor reports the energy present within its associated frequency band, and the combined inputs from multiple such sensors follow a single sensor input trajectory, and the distribution of activity levels across all nodes is used to identify the speaker and / or the spoken letters, syllables, words, phrases, or phonemes present in the audio signal.

[0014] According to a second embodiment, a computer program product is provided which includes a non-temporary computer-readable medium on which a computer program including program instructions is provided, the computer program is loadable into a data processing unit, and the computer program is configured to cause the data processing unit to execute the method or one of the embodiments described above when the computer program is executed by the data processing unit.

[0015] According to a third embodiment, an entity identification device is provided, which includes a control circuit configured to: a) supply inputs from multiple sensors to a network of multiple nodes; b) generate an activity level at each node of the network based on the inputs from the multiple sensors; c) compare the activity level of each node with a threshold level; d) set the activity level of each node to a preset value or retain the generated activity level based on the comparison step; e) calculate a total activity level as the sum of all activity levels of the nodes in the network; f) repeat steps a) to e) until a local minimum of the total activity level is reached; and g) once a local minimum of the total activity level is reached, utilize the distribution of activity levels at the local minimum to identify measurable entity characteristics.

[0016] Further effects and features of the second and third embodiments are largely similar to those described earlier with reference to the first embodiment, and vice versa. Embodiments described in relation to the first embodiment are largely compatible with the second and third embodiments, and vice versa.

[0017] One advantage of some embodiments is that they provide an alternative approach to entity identification.

[0018] One advantage of some embodiments is that improved entity identification performance is achieved.

[0019] Another advantage of some embodiments is that they provide more reliable entity identification.

[0020] An advantage of some embodiments is that the device can be trained much faster, for example, because the device is more generalizable or more adaptable, for example, due to improved dynamic performance.

[0021] Another advantage of some embodiments is that training the processing elements is much faster, for example, because they require only a small training dataset.

[0022] Another advantage is that the device can self-train, meaning that a limited amount of initial training data is infused into the device, and its representation is supplied as new combinations via the network, thereby enabling a kind of "data enhancement." However, in this case, this is not finely tuned sensor data, but rather an internal representation of perceptual information being played back to the device, resulting in more efficient self-training than would be achieved through data enhancement alone.

[0023] A further advantage of some embodiments is that they provide an efficient or even more efficient method for identifying entities.

[0024] Another further advantage of some embodiments is that they provide an energy-efficient entity identification method, for example, because this method saves power and / or storage space on the computer.

[0025] Another further advantage of some embodiments is that they provide a bandwidth-efficient method for identifying a portion of information, for example, because this method saves bandwidth required to transmit the data.

[0026] The details of this disclosure will become apparent from the following detailed description. The detailed description and specific examples are merely illustrative of preferred embodiments of this disclosure. Modifications and alterations can be made within the scope of this disclosure as will be obvious to those skilled in the art from the guidance in the detailed description.

[0027] Therefore, it should be understood that the disclosures disclosed herein are not limited to any particular component part of the steps of the described apparatus or method, for such apparatus and method may vary. Similarly, it should be understood that the terminology used herein is for the purpose of describing a particular embodiment only and is not intended to be limiting. It should be noted that, as used herein and in the appended claims, indefinite articles, definite articles and the foregoing are intended to mean the presence of one or more of the elements unless otherwise expressly indicated by the context. Thus, for example, a reference to “one unit” or “this unit” may include multiple devices and similar ones. Furthermore, the words and similar phrases “have,” “include,” and “contain” are not intended to exclude other elements or steps.

[0028] Regarding technical terms, the term "measurable" should be interpreted as something that can be measured or detected, i.e., detectable. The terms "measure" and "detect" should be interpreted as synonyms. The term "entity" should be interpreted as an entity, such as a physical entity, or as a more abstract entity, such as a financial entity, e.g., one or more financial datasets. The term "physical entity" should be interpreted as an entity that has a physical existence, e.g., an object, (object's) features, gestures, applied pressure, speaker, uttered letters, syllables, phonemes, words, or phrases. The term "node" can be a neuron (in a neural network) or other processing element.

[0029] The following exemplary and non-limiting detailed description of exemplary embodiments of the present disclosure, when read in conjunction with the accompanying drawings, will enable a more complete understanding of the above problems as well as additional problems, features, and advantages of the present disclosure.

Brief Description of the Drawings

[0030] [Figure 1] It is a flowchart showing exemplary method steps according to some embodiments of the present disclosure. [Figure 2] It is a schematic diagram showing an exemplary computer-readable medium according to some embodiments. [Figure 3] It is a schematic block diagram showing an exemplary device according to some embodiments. [Figure 4] It is a schematic diagram showing the operating principle of a device according to some embodiments, using one example including a plurality of sensors and a plurality of processing elements. [Figure 5] It is a schematic diagram showing the operating principle of a device according to some embodiments, using one example including a plurality of sensors and a plurality of processing elements. [Figure 6] It is a schematic diagram showing the operating principle of a device according to some embodiments, using one example including a plurality of processing elements. [Figure 7] It is a schematic diagram showing the operating principle of a device according to some embodiments, using one example including a plurality of sensors and a plurality of processing elements. [Figure 8] Figures 8A to 8H are schematic diagrams showing the operating principle of a device according to some embodiments, using one example including a plurality of tactile sensors. [Figure 9] It is a schematic diagram showing the operating principle of a device according to some embodiments, using one example including a camera. [Figure 10] Figures 10A to 10C are graphs showing the frequencies and powers related to various sensors for detecting audio signals. [Figure 11] It is a graph of a perception trajectory.

[0031] Detailed explanation Next, the Disclosure will be described with reference to the accompanying drawings illustrating preferred exemplary embodiments thereof. However, the Disclosure may be carried out in other forms, and the Disclosure should not be construed as being limited to the embodiments disclosed herein. The disclosed embodiments are provided to fully convey the scope of the Disclosure to those skilled in the art.

[0032] Embodiments are described below, with Figure 1 being a flowchart illustrating exemplary method steps according to one embodiment of the present disclosure. Figure 1 shows an entity identification method 100 implemented by computer or hardware. Thus, this method can be implemented as hardware, software, or any combination of the two. The method includes step 110 of supplying inputs 502, 504, 506, 508 from a plurality of sensors to a network 520 (shown in Figure 5) consisting of nodes 522, 524, 526, 528. The network 520 can be a recurrent network, such as a recurrent neural network. The sensors can be any suitable sensors, such as image sensors (e.g., pixels), audio sensors (e.g., microphones), or tactile sensors (e.g., pressure sensor arrays or biologically inspired tactile sensors). Furthermore, these inputs can be generated in operation, i.e., the sensors are directly connected to the network 520. Alternatively, the inputs are first generated and recorded / stored and then supplied to the network 520. Furthermore, the inputs can be preprocessed. One way to preprocess the input is to combine or recombine multiple sensor signals (for example, by averaging or adding the signals) and use such combined or recombined signals as one or more inputs to the network 520. In either case, the input data may change continuously over time and progress gradually. Furthermore, this method includes step 120, in which each node 522, 524, 526, 528 of the network 520 generates an activity level based on inputs from multiple sensors. Thus, each node 522, 524, 526, 528 has an activity level associated with it. The activity level of one node represents the state of that node, for example, its internal state. Furthermore, the activity level set, i.e., the activity levels of each node, represents the internal state of the network 520.The internal state of network 520 can be used as a feedback signal to one or more actuators, which directly or indirectly control the input, thereby controlling the sensor input trajectory. Furthermore, this method includes a step 130 which compares the activity level of each node 522, 524, 526, 528 with a threshold level, i.e., an activity level threshold. The threshold level can be any appropriate value. Based on the comparison step 130, this method includes a step 140 which sets the activity level of each node 522, 524, 526, 528 to a preset value or retains the generated activity level. According to some embodiments, if the generated activity level is higher than the threshold level, that level is retained, and if the generated activity level is less than or equal to the threshold level, the activity level is set to a preset value. The activity level can be set to a preset value of zero or any other appropriate value. Furthermore, this method includes a step 150 which calculates the total activity level as the sum of all activity levels of nodes 522, 524, 526, 528 of network 520. For this calculation, the set value is used along with any retained generated values. If the preset value is zero, the total activity level can be calculated as the sum of only all retained generated values, thus ignoring all zero values ​​and resulting in a faster calculation. Furthermore, this method includes step 160, which repeats steps 110, 120, 130, 140, and 150 described above until a local minimum of the total activity level is reached. This iteration is performed by a continuously progressing input signal, for example, each input being a continuous signal from a sensor. The local minimum is reached when the lowest possible total activity level is achieved. If the total activity level falls below the total activity threshold for a predetermined number of iterations, it can be considered that the lowest possible total activity level has been reached. This number of iterations can be any appropriate number, such as 2.If the total activity level reaches a local minimum, the activity level distribution at the local minimum is used to identify one (or more) measurable characteristics of the entity (step 170). Measurable characteristics can be features of an object, parts of features, positional trajectories, applied pressure trajectories, or the frequency discrimination characteristics of a particular speaker when they utter a particular letter, syllable, phoneme, word, or phrase. In this case, such measurable characteristics can be mapped to entities. For example, a feature of an object can be mapped to an object, a part of a feature can be mapped to a feature (of an object), a positional trajectory can be mapped to a gesture, an applied pressure trajectory can be mapped to the (maximum) applied pressure, the frequency discrimination characteristics of a particular speaker can be mapped to that speaker, and uttered letters, syllables, phonemes, words, or phrases can be mapped to actual letters, syllables, phonemes, words, or phrases. Such mapping can simply be a lookup in memory, a lookup table, or a database. This search can be based on finding the entity among multiple physical entities that has the characteristics most closely matching the identified measurable characteristics. From such a search, the actual entity can be identified. Once a local minimum is reached, network 520 traverses the local minimum trajectory for a user-definable time. The accuracy of the identification depends on the total time spent traversing this local minimum trajectory.

[0033] According to some embodiments, the input changes dynamically over time and follows a single (temporal) sensor input trajectory or perception trajectory. According to some embodiments, multiple sensors monitor dependencies between sensors. Such dependencies may result from the fact that multiple sensors are located on different parts of the same substrate, for example, that these sensors are located close to each other and therefore measure different situations of a single signal that is related or dependent on each other. Alternatively, the quantities that sensors measure may have dependencies due to the laws governing the world monitored by the sensors. For example, if the visual world is mapped (by a camera) to a set of sensor pixels in a first image, two adjacent sensors / pixels may have high contrast in their brightness, but not all adjacent sensors / pixels will have high intensity contrast between them, because the visual world is not constructed in that way. For this reason, there may be some degree of predictability or dependency in the visual world, meaning the following: In other words, if there is high brightness contrast between the central pixel and the first pixel, for example, the pixel to the left of the central pixel, but much lower brightness contrast between the central pixel and other adjacent pixels (for example, the pixels to the right, above, or below the central pixel), then this relationship may be reflected in other images, such as the second image. That is, in the second image as well, there may be high brightness contrast between the central pixel and the first pixel, while there may be much lower brightness contrast between the central pixel and other adjacent pixels.

[0034] In some embodiments, the local minimum of the total activity level is reached when a single sensor input trajectory is followed for a period longer than a user-definable time threshold and with a deviation smaller than a user-definable deviation threshold. In other words, the local minimum of the total activity level is reached when the sensor input trajectory is followed sufficiently well for a sufficient period of time by the internal trajectory, which is the trajectory followed by nodes 522, 524, 526, and 528. Therefore, in some embodiments, the sensor input trajectory is reproduced or represented by the internal trajectory. Furthermore, the local minimum is reached when the total activity level is as low as possible, that is, when the total activity level (of network 520) relative to the sum of the activities supplied to network 520 by inputs from multiple sensors is as low as possible. Since the deviation threshold and time threshold are user-definable, the user can select an appropriate precision. Furthermore, since the deviation threshold and time threshold are user-definable, it is not necessary to reach or find an actual local minimum; instead, it is sufficient that the total activity level is near or close to the local minimum, depending on the set deviation threshold and time threshold, and thus deviations from the internal trajectory are permitted. Moreover, if the total activity level is not within the range of the set deviation threshold and time threshold, this method optionally includes a step of reporting that the entity cannot be identified with adequate accuracy / certainty.

[0035] According to some embodiments, the computer program product includes a non-temporary computer-readable medium 200, such as a Universal Serial Bus (USB) memory, a plug-in card, an embedded drive, a Digital Multipurpose Disc (DVD), or a read-only memory (ROM). Figure 2 shows an exemplary computer-readable medium in the form of a Compact Disc (CD) ROM 200. The computer-readable medium stores a computer program containing program instructions. The computer program can be loaded into a data processor (PROC) 220, which may be included, for example, in a computer or computing device 210. Once loaded into the data processing unit, the computer program can be stored in memory (MEM) 230 associated with or included in the data processing unit. According to some embodiments, once the computer program is loaded into the data processing unit and executed by the data processing unit, it can perform method steps according to the method shown in Figure 1, as described herein.

[0036] Figure 3 is a schematic block diagram showing an exemplary apparatus according to some embodiments. Figure 3 shows an entity identification apparatus 300. The apparatus 300 can be configured to perform (e.g., carry out) one or more method steps, such as those shown in Figure 1 or described in other embodiments herein. The apparatus 300 includes a control circuit 310, which is configured as follows: Specifically, the network 520, consisting of nodes 522, 524, 526, and 528, is supplied with inputs from multiple sensors (see step 110 in Figure 1). Based on the inputs from the multiple sensors, each node 522, 524, 526, and 528 of the network 520 generates an activity level (see step 120 in Figure 1). The activity level of each node 522, 524, 526, and 528 is compared with a threshold level (see step 130 in Figure 1). Based on the comparison, the activity level of each node 522, 524, 526, and 528 is set to a preset value, or the generated activity level is retained. (See step 140 in Figure 1) The system is configured to calculate the total activity level as the sum of all activity levels of nodes 522, 524, 526, and 528 of network 520 (See step 150 in Figure 1), and to repeat supplying, generating, comparing, setting / holding, and calculating until the total activity level reaches a local minimum (See step 160 in Figure 1). Furthermore, once the total activity level reaches a local minimum, the system is configured to utilize the distribution of activity levels at the local minimum to identify measurable entity characteristics (See step 170 in Figure 1).

[0037] The control circuit 310 may include the following components, or the control circuit 310 may be associated with the following components: namely, a supply device (e.g., a supply circuit or supply module) 312 configured to supply inputs from multiple sensors to a network consisting of multiple nodes; a generator (e.g., a generation circuit or generation module) 314 configured to generate an activity level by each node of the network based on the inputs from multiple sensors; a comparator (e.g., a comparator circuit or comparator module) 316 configured to compare the activity level of each node with a threshold level; and a setting / holding device (e.g., a setting / holding circuit or...) configured to set the activity level of each node to a preset value or to hold the generated activity level based on the comparison. A configuration including a setting / holding module 318, a computing device (e.g., a computing circuit or computing module) 320 configured to calculate the total activity level as the sum of all activity levels of the network nodes, an iterative device (e.g., an iterative circuit or iterative module) 322 configured to repeatedly supply, generate, compare, set / hold and calculate until a local minimum of the total activity level is reached, and a utilization device (e.g., a utilization circuit or utilization module) 324 configured to utilize the distribution of activity levels at the local minimum to identify measurable entity characteristics once the local minimum of the total activity level has been reached.

[0038] Figure 4 illustrates the operating principle of a device according to one embodiment, using one example that includes multiple sensors and multiple nodes / processing elements. In Figure 4-1, a network 420 including nodes i1, i2, i3, and i4 is activated by activation energy X. According to some embodiments, as schematically shown in Figure 4, the network consisting of nodes i1, i2, i3, and i4 has nonlinear attractor dynamics. More specifically, the network consisting of multiple nodes is an attractor network. Furthermore, nonlinearity is introduced by setting the activity level to a preset value based on a comparison 130 between the activity level of each node and a threshold level. The internal state of network 420 is equal to the distribution of activity across nodes i1 to i4. The internal state progresses continuously and gradually over time according to the structure of network 420. In Figure 4-2, the internal state of network 420 is used to trigger motions that generate sensor activation or simply to perform matrix operations on sensor data. In some embodiments, as shown in Figure 4, network 420 generates asynchronous data, which in turn triggers sensor activation. In Figure 4-3, the external state of the surrounding world, schematically shown in Figure 4 as an example, is measured by sensors j1-j4 in sensor network 430 as an object detected by multiple sensors, such as biotouch sensors. In some embodiments, the relationship between the sensors and the outside world can be changed, for example, if the internal state of network 420 is used to trigger a motion that generates sensor activation (a change in sensor activation caused by actuator activation due to an internal state). Figure 4-4 shows, for example, sensor network 430 that generates asynchronous data supplied to network 420. Sensors j1-j4 of sensor network 430 are always dependent on visual or audio signals, mechanically and / or due to physical characteristics of the outside world, and their dependencies can be compared to a network with state-dependent weights.The activation energy X will influence where the local minimum of the total activity level lies. Furthermore, the activation energy X can also drive the output from i1-i4 to j1-j4 (or actuators) (the distribution of activity levels across all nodes). Thus, the activation energy X helps to make the sensor input trajectory a function of the internal state, for example, a function of the internal trajectory. The activation energy X can be an initial estimate or prediction of a particular entity, or a request for information / a specific part of an entity for a given sensing condition known to consist of a combination of many entities.

[0039] Figure 5 illustrates the operating principle of a device according to one embodiment, which includes multiple sensors and multiple processing elements or nodes. More specifically, Figure 5 shows that each node of a network 520, consisting of nodes 522, 524, 526, and 528, has individual inputs 502, 504, 506, and 508 from individual sensors (not shown).

[0040] Figure 6 is a schematic diagram illustrating the operating principle of a device according to one embodiment, which includes multiple interconnected nodes or processing elements. Figure 6 shows a network 520 consisting of nodes 522, 524, 526, and 528, where each node 522, 524, 526, and 528 is connected to all other nodes 522, 524, 526, and 528. If all nodes 522, 524, 526, and 528 are connected to all other nodes 522, 524, 526, and 528, a system / method with maximum potential variance can be obtained. Thus, the richness of potential representation becomes possible. According to this scheme, each node can be added to improve the richness of representation. To achieve this, the precise distribution of connection lines / weights between nodes is a crucial tolerance factor.

[0041] Figure 7 is a schematic diagram illustrating the operating principle of a device according to one embodiment, using one example that includes multiple sensors and multiple nodes / processing elements. Figure 7 shows a network 520 consisting of nodes 522, 524, 526, and 528, each of which is connected to all other nodes 522, 524, 526, and 528 via connecting lines. Furthermore, each node 522, 524, 526, and 528 is provided with at least one input, such as inputs 512, 514 from multiple sensors. According to one embodiment, the activity level of each node 522, 524, 526, and 528 is used as an input via the connecting lines, and each input is weighted by a weight (input load, e.g., synaptic load) relative to all other nodes 522, 524, 526, and 528. At least one weighted input is negative. One way to achieve this is by utilizing at least one negative weight. Alternatively or additionally, at least one weighted input is positive. One way to achieve this is by utilizing at least one positive weight. According to one embodiment, some nodes 522,524 influence all other nodes with weights having values ​​between 0 and +1, while other nodes 526,528 influence all other nodes with weights having values ​​between -1 and 0. Alternatively or additionally, all retained generating activity levels are positive scalars. By combining the use of negative weights at any given time with preset values ​​for all retained generating activity levels, which are positive scalars, and all other activity levels, which are, for example, zero, some nodes (which were not below the threshold level at a preceding point in time) can be made below the threshold level for generating an output, which means that the effective structure of the network can be dynamically changed during the identification process.According to this embodiment, this method differs from the method using Hopfield networks not only in that a threshold is applied to the activity level of each node, but also because, although the nodes in this embodiment only have positive scalar outputs, they can generate negative inputs. Furthermore, if all nodes / neurons are interconnected and all inputs from the sensor target all nodes / neurons in the network, the richness of representation can be achieved. In this case, the input from the sensor induces various states in the network, depending on the precise spatiotemporal pattern of the sensor input (and the time evolution of that pattern).

[0042] As explained earlier with reference to Figure 1, the local minimum is reached when the total activity level is as low as possible, that is, when the total activity level (of network 520) is as low as possible relative to the sum of the activities supplied to network 520 by inputs from multiple sensors. However, when the inputs from multiple sensors are the product of the activities of nodes 522, 524, 526, and 528 of network 520 (and when nodes 522, 524, 526, and 528 of the network are driven by activation energy X), the most effective solution is simply to set all input loads to zero. In fact, each node 522, 524, 526, and 528 of network 520 also has a drive mechanism that actively avoids all input loads becoming zero. More specifically, according to some embodiments, each node 522, 524, 526, and 528 of network 520 has means / mechanisms to prevent all input loads of those nodes from becoming zero. Furthermore, according to some embodiments, the network 520 has additional means / mechanisms to prevent the weights of all perceptual inputs from becoming zero.

[0043] Figure 8 is a schematic diagram illustrating the operating principle of a device equipped with multiple tactile sensors. More specifically, Figure 8 shows how the same type of sensor dependency defining a single dynamic feature can arise under two different detection conditions: touch or stretch on a rigid surface and touch or stretch on a flexible surface. According to some embodiments, the multiple sensors are touch sensors or tactile sensors. The tactile sensors can be, for example, a pressure sensor array or a biologically inspired tactile sensor. Each tactile sensor in an array, for example, detects whether a surface has been touched or stretched, and whether it has been activated by the touch / stretch, for example, by a finger, pen, or other object. The tactile sensors are positioned where the finger, pen, or other object would be. When a tactile sensor is activated, it outputs a touch event signal, e.g., +1. If a tactile sensor is not activated, it outputs a non-touch event signal, e.g., 0. Alternatively, if a tactile sensor detects that the surface is being touched, the sensor outputs a touch event signal with a force-dependent value, e.g., a value between 0 and +1; if the tactile sensor does not detect that the surface is being touched, the sensor outputs a non-touch event signal, e.g., 0. The output of the tactile sensor is supplied as input to a network 520 consisting of nodes 522, 524, 526, and 528. Thus, the input from each of the multiple sensors is either a touch event (e.g., with a force-dependent value) or a non-touch event. According to some embodiments, the surface being touched / stretched is a rigid (non-flexible) surface. For example, when a finger first touches the surface, as shown in Figure 8A, only one or a few tactile sensors in the array may detect this as a touch event. Then, when a flexible finger is pressed onto the surface with a certain force, the contact involves a larger surface area, thereby causing more sensors to detect the touch event, as shown in Figures 8B-8D.As shown in Figures 8A to 8D, if threshold levels are selected (using the method described with reference to Figure 1) so that only the activity levels of nodes 522, 524, 526, and 528, which have input from the tactile sensors receiving the highest shear force, are retained, then only the sensors in the intermediate zone represented by the edges / boundaries of the circles in Figures 8A to 8D are retained. Thus, as a finger is pressed against the surface with a constant force, the contact gradually involves a larger surface area, and the retained time-dependent generated activity level can be represented as a wave moving radially outward; that is, the traced sensor input trajectory is a wave moving radially outward that is involved in a predictable sensor activation sequence. As the finger is lifted, and therefore the finger gradually ceases to touch the surface, the traced sensor input trajectory is a wave moving radially inward across the cluster of skin sensors. These trajectories can be used to identify new contact events and / or the end of contact events. For example, these trajectories can be used to distinguish between different types of contact events by comparing them to the trajectories of known contact events. Alternatively, several adjacent trajectory paths / trajectory components are added, which may or may not be discovered by a system that depends on a threshold for identification, even though both can follow the same entire trajectory. By using trajectories for identification, new contact events (or the end of a contact event) can be identified as the same type of spatiotemporal sequence or qualitative event, regardless of the amount of force applied by the finger and the absolute level of the resulting shear force, i.e., regardless of how fast or slow the finger is applied to the surface (in this case, the speed of the finger movement may also depend, for example, on the activation energy X mentioned above). Furthermore, the identification does not depend on whether one or a few sensors are malfunctioning / interfering, which leads to robust identification.

[0044] According to some embodiments, a tactile sensor or tactile sensor array touches / stretches a flexible surface. The flexible surface causes the intermediate zone to expand instead, as shown in Figures 8E-8H, widening the intermediate zone. However, the overall sensor activation relationship (sensor input trajectory) remains the same, and this method, provided the threshold level is set to a sufficiently acceptable / low level, results in the total system activity ending at the same local minimum, allowing the characteristics of the contact action (new contact event and / or termination of the contact event) to be identified. Alternatively, the distribution of activity levels across all nodes can be used to identify the sensor input trajectory as a gesture. For example, a rigid surface can be used to identify a gesture, while a flexible surface can be used to identify, for example, that a maximum shear force is being applied at a given interval.

[0045] Figure 9 is a schematic diagram illustrating the operating principle of a device with a camera. According to some embodiments, the inputs from multiple sensors are pixel values. Pixel values ​​can be luminance values. Alternatively or additionally, pixel values ​​can be luminance values ​​for one or more components representing colors such as red, green, and blue, or cyan, magenta, yellow, and black. Pixel values ​​can be part of an image captured by a camera 910 (shown in Figure 9), such as a digital camera. Furthermore, the image can be multiple images captured in a single sequence. Alternatively, the image can be a subset of captured images, such as every other image in a sequence. Using the method described with reference to Figure 1, the sensor input trajectory can be controlled by controlling the position of the camera 910 through rotational and / or translational motion of the camera 910, utilizing the distribution of activity levels across all nodes 522, 524, 526, and 528. Rotational and / or translational motion can be performed by one or more actuators 912, such as motors, configured to rotate / angle-adjust the camera and / or move the camera forward, backward, or left and right. Therefore, the distribution of activity levels across all nodes 522, 524, 526, and 528 is used as a feedback signal to actuator 912. As shown in Figure 9, camera 910 has an object 920 in its focal region. Thus, object 920 will be present in one or more of the captured images. Object 920 can be a person, a tree, a house, or any other suitable object. By controlling the angle or position of the camera, input from multiple pixels is affected / modified. In this case, the sensor signal, i.e., the pixel, becomes a function of the distribution of activity levels across all nodes 522, 524, 526, and 528, i.e., a function of the distribution of its own internal state. The active movement of camera 910 generates a time-evolving stream of sensor input. Thus, the input changes dynamically over time and follows a single sensor input trajectory.The sensor input trajectory is controlled by the movement of the camera 910, and when it reaches a local minimum of the total activity level, the distribution of activity levels at this local minimum is used to identify measurable features of an entity, such as the features of an object or a part of its features. In this case, if the measurable feature is a feature of an object, the entity is an object; if the measurable feature is a part of a feature, the entity is a feature. The feature can be a biometric feature, such as the distance between two biometric points, such as the distance between the eyes of one person. Alternatively, the feature can be the width or height of an object, such as the width or height of a tree. According to some embodiments, as the distance between the object 920 and the camera 910 increases, the number of pixels used as input can be reduced, and vice versa, thereby ensuring that objects or features of the same object are identified as the same entity. For example, if the distance between object 920 and camera 910 is doubled, only a quarter of the pixels used as input at shorter distances will be used, meaning that pixels covering the object at longer distances will be used as input at even longer distances. Thus, objects / entities can be identified as the same object / entity regardless of distance. This is because the sensor dependencies when the camera sweeps across an object remain qualitatively the same and identifiable even when the object is located further away, despite the smaller number of sensors / pixels involved. In another example, the feature to be identified is the vertical contrast line between the white and black fields. If camera 910 sweeps across the vertical contrast line (for example, from left to right), the four sensors will find the same feature (i.e., the vertical contrast line) as 16 or 64 sensors. Therefore, there exists a central feature element that becomes a specific type of sensor activation dependency that moves across multiple sensors as the camera moves. Furthermore, the camera's motion speed can also be controlled.When camera 910 is moving at an increased speed, the pixels used as input can be taken from even fewer images, such as from every other image, and vice versa, thereby ensuring that objects or features of objects are identified as the same elements regardless of speed. Furthermore, entity identification and / or identification of measurable entity characteristics can be associated with tolerance thresholds; that is, tolerance thresholds can be used to determine whether a match exists when comparing known physical entities or their characteristics with current physical entities or their characteristics in memory, lookup tables, or databases (using the activity level distribution at the found local minimum). Tolerance thresholds can be set by the user, i.e., they can be user-definable. By using tolerance thresholds, it is ensured that features or objects can be identified as the same even if the activity distribution across multiple sensors is not exactly the same. To determine whether a match exists, the activity distribution across multiple sensors does not need to be exactly the same. Therefore, even if one or a few sensors / pixels are not functioning properly or are interfering, correct identification can still be achieved; in other words, there is relative robustness to interference. Thus, features or objects can be identified at close range / short distances, but the same features can also be identified at longer distances; that is, features or objects can be identified regardless of distance. The same logic applies to two objects that are at the same distance, different in size, but have the same features. In both cases, the total sensor activity changes, but their overall spatiotemporal activation relationship can still be recognized.

[0046] Figures 10A to 10C show the frequencies and powers for various sensors that detect audio signals. As can be seen from Figure 10A, the frequency spectrum of an audio signal may be divided into various frequency bands. Power or energy in these various frequency bands can be detected (and reported) by the sensors. Figure 10B shows the power in the frequency bands detected by sensors 1 and 2. As can be seen from this figure, each sensor detects various frequency bands. Figure 10C shows the perceptual trajectory based on the power detected over time by sensors 1 and 2. An audio signal may contain sounds from a single voice (and possibly other sounds) that include dynamic changes in power or energy across multiple frequency bands. Thus, for each uttered syllable, a voice can be identified as belonging to a specific individual (of a group of individuals) based on specific dynamic identification characteristics. Dynamic identification characteristics include specific changes in power or energy levels in each frequency band over a given period of time. Therefore, by dividing the frequency spectrum into frequency bands and providing sensors (one or more sensors per frequency band) that detect power or energy in each of the frequency bands or multiple frequency bands, the dynamic identification characteristics generate a specific sensor input trajectory. Thus, the combined inputs from multiple such sensors follow a single sensor input trajectory. According to some embodiments, each sensor of the multiple sensors is associated with a frequency band of a single audio signal. Preferably, each sensor is associated with a different frequency band. Each sensor detects (and reports) the power or energy present in the frequency band associated with each sensor over a predetermined period of time. Using the method described with reference to Figure 1, once a local minimum is reached, the distribution of activity levels across all nodes 522, 524, 526, 528 at that local minimum is used to identify the speaker.Alternatively or additionally, the activity levels across all nodes 522, 524, 526, and 528 can be used to identify spoken letters, syllables, phonemes, words, or phrases present in the audio signal. For example, a syllable can be identified by comparing the distribution of activity levels across all nodes 522, 524, 526, and 528 at a found local minimum with a stored distribution of activity levels associated with a known syllable. Similarly, a speaker can be identified by comparing the distribution of activity levels across all nodes 522, 524, 526, and 528 at a found local minimum with a stored distribution of activity levels associated with a known speaker. To determine whether a match exists (for a syllable or speaker), tolerance thresholds, as described with reference to Figure 9, can also be applied. Moreover, this identification is independent of the velocity and volume of the acoustic signal, as a trajectory is followed to reach a local minimum for identification.

[0047] Figure 11 is a graph of another sensor input trajectory. Figure 11 shows a single sensor input trajectory over time based on three sensors (sensor 1, sensor 2, and sensor 3). As can be seen from Figure 11, the measured values ​​from sensor 1 over time are used as the X coordinate, the measured values ​​from sensor 2 over time are used as the Y coordinate, and the measured values ​​from sensor 3 over time are used as the Z coordinate in a Cartesian coordinate system for three-dimensional space. According to one embodiment, these sensors measure different frequency bands of a single audio signal. According to another embodiment, the sensors measure whether a touch event is present. According to yet another embodiment, the sensors measure the brightness value of a pixel. The coordinates depicted in Figure 11 combine to form a single perceptual trajectory.

[0048] As will be obvious to those skilled in the art, this disclosure is not limited to the preferred embodiments described above. As will be even more obvious to those skilled in the art, modifications and alterations can be made within the scope of the appended claims. For example, other entities such as aromas or flavors can be identified. In addition to these, those skilled in the art will be able to understand and achieve alterations to the disclosed embodiments by examining the drawings, disclosures, and appended claims in carrying out the claimed disclosure.

Claims

1. A method (100) for entity identification implemented by computer or hardware, wherein the method (100) is a) Step (110) of supplying inputs from multiple sensors to a network consisting of multiple nodes, b) A step (120) in which each node of the network generates an activity level based on the inputs from the plurality of sensors, c) A step (130) of comparing the activity level of each node with a threshold level, d) Based on the comparison step, for each node, the step (140) of setting the activity level to a preset value or retaining the generated activity level, e) A step (150) of calculating the total activity level as the sum of all activity levels of the nodes in the network, f) Step (160) of repeating steps a) to e) until the total activity level reaches a local minimum, g) If the total activity level reaches the local minimum, the step (170) is to use the distribution of activity levels at the local minimum to identify a measurable characteristic of the entity, Includes, The aforementioned input changes dynamically over time, following a single sensor input trajectory. The state in which the total activity level reaches its local minimum occurs when a single sensor input trajectory is followed for a period longer than a user-definable time threshold and with a deviation smaller than a user-definable deviation threshold. A method for entity identification implemented by a computer or hardware (100).

2. The method, implemented by a computer or hardware according to claim 1, wherein the plurality of sensors monitor the dependencies between the sensors.

3. A computer or hardware-implemented method according to claim 1 or 2, wherein if the total activity level falls below a total activity threshold for a predetermined number of iterations, it is determined that the local minimum value of the total activity level has been reached.

4. The method further includes, when the total activity level reaches the local minimum, The steps include: utilizing the distribution of the total activity level at the local minimum to identify a measurable characteristic or a set of measurable characteristics of the entity; The steps include mapping an identified measurable characteristic or a set of identified measurable characteristics to the entity, A method implemented by a computer or hardware according to any one of claims 1 to 3, including the following:

5. The computer or hardware-implemented method according to any one of claims 1 to 4, wherein the entity is a physical entity and includes one of an object, an object feature, a gesture, applied pressure, a speaker, uttered letters, syllables, phonemes, words, or phrases.

6. A computer or hardware-implemented method according to any one of claims 1 to 5, wherein the activity level of each node is used for all other nodes as a weighted input, each weighted input being at least one negative and / or at least one positive and / or all retained generating activity levels being positive scalars.

7. The computer or hardware-implemented method according to any one of claims 1 to 6, wherein the inputs from the plurality of sensors are pixel values, e.g., brightness, of an image captured by a camera, and the distribution of activity levels across all nodes is further used to control the position of the camera by rotational and / or translational motion of the camera, thereby controlling the sensor input trajectory, and the entities identified are objects or features of objects present in at least one of the captured images.

8. A computer or hardware-implemented method according to any one of claims 1 to 6, wherein the plurality of sensors are touch sensors, the input from each of the plurality of sensors is a touch event signal having a force-dependent value, and the distribution of activity levels across all nodes is used to identify the sensor input trajectory as a new contact event, the end of a contact event, a gesture, or applied pressure.

9. A computer or hardware-implemented method according to any one of claims 1 to 6, wherein each of the plurality of sensors is associated with a different frequency band of a single audio signal, each sensor reports the energy present within the associated frequency band, the combined inputs from the plurality of such sensors follow a single sensor input trajectory, and the distribution of activity levels across all nodes is used to identify the speaker and / or the spoken letters, syllables, phonemes, words or phrases present in the audio signal.

10. A recording medium on which a computer program including program instructions is recorded, wherein the computer program is loadable into a data processing unit (220), and the data processing unit (220) is configured to execute the method described in any one of claims 1 to 9 when it executes the computer program.

11. An entity identification device (300) includes a control circuit (310) configured to perform the following: a) A network consisting of multiple nodes is supplied with inputs from multiple sensors. b) Each node of the network generates an activity level based on the inputs from the multiple sensors, c) Compare the activity level of each node with the threshold level, d) Based on the comparison, set the activity level for each node to a preset value, or retain the generated activity level. e) The total activity level is calculated as the sum of all activity levels of the nodes in the network, f) Repeat steps a) to e) until the total activity level reaches its local minimum value. g) If the total activity level reaches the local minimum, use the activity level distribution at the local minimum to identify the measurable characteristics of the entity. It includes a control circuit (310) configured to perform the action, The aforementioned input changes dynamically over time, following the sensor input trajectory. The state in which the total activity level reaches its local minimum occurs when a single sensor input trajectory is followed for a period longer than a user-definable time threshold and with a deviation smaller than a user-definable deviation threshold. Entity identification device (300).

Images