Calibration of an electrochemical sensor to generate an embedding in the embedding space.

Abstract

The present invention provides a computing system for calibrating an electrochemical sensor to generate an embedding in a suitable embedding space. [Solution] The system 400 includes a sensor that generates sensor data indicating the presence of compounds in the environment, and an embedding model that receives the sensor data and is trained to generate an embedding in the embedding space, the embedding model including ground truth characteristic labels that are applied to a set of electrical signals generated by one or more test sensors when exposed to a plurality of training compounds, each ground truth characteristic label being trained by a training set that describes the characteristics of one or more training compounds, and the system processes the generation of sensor data generated by the sensor using the embedding model to generate an output.

Description

Technical Field

[0001] Cross-Reference to Related Applications This application claims priority and benefit to U.S. Provisional Patent Application No. 63 / 189,501, filed on May 17, 2021. The U.S. Provisional Patent Application No. 63 / 189,501 is incorporated herein by reference in its entirety.

[0002] The present disclosure generally relates to processing sensor data for detecting and / or generating representations of chemical molecules. More specifically, the present disclosure relates to generating sensor data, processing the sensor data with a machine-learned model to generate an embedding output, and performing various tasks using the embedding output.

Background Art

[0003] Although computing devices can be used for visual computing or audio processing, computing devices cannot reliably detect odors. While chemical sensors are available, they generate raw signals that are difficult to interpret. Chemical sensors cannot convert raw signals into human-interpretable labels such as "orange" or "cinnamon" across the entire space of possible odors. Some computing devices are configured to determine a small subset of odors based on personal training, but these computing devices cannot determine properties for which they have not been trained.

[0004] Furthermore, individually training for all possible odors is time-consuming and computationally expensive after the final configuration is completed, and even after such training, combinations of known odors cannot be identified. An odor is only associated with input data, making it impossible to determine the olfactory properties of new mixtures.

Summary of the Invention

Means for Solving the Problems

[0005] Aspects and advantages of the embodiments of this disclosure are partially described in the following description, can be learned from the description, or can be learned through the practice of the embodiments.

[0006] One exemplary aspect of the present disclosure relates to a computing system. The computing system may include sensors configured to generate electrical signals indicating the presence of one or more compounds in an environment, and a machine learning model trained to receive and process the electrical signals to generate an embedding in an embedding space. In some embodiments, the machine learning model may be trained using a training dataset that includes multiple training examples, each training example including a ground truth characteristic label applied to a set of electrical signals generated by one or more test sensors when exposed to one or more training compounds. Each ground truth characteristic label may describe the characteristics of one or more training compounds. The computing system may include one or more processors and one or more non-temporary computer-readable media that, when executed by one or more processors, collectively store instructions causing the computing system to perform actions, which may include generating sensor data by sensors indicating the presence of a particular compound in an environment, and processing the sensor data by one or more processors using a machine learning model to generate an embedding output into an embedding space.

[0007] In some embodiments, the operation may include performing a task based on the embedded output. The task may include providing predictions of sensory characteristics based on the embedded output. In some embodiments, the task may include providing predictions of olfactory characteristics based on the embedded output. The task may be to identify a disease state based at least partially on the embedded output. In some embodiments, the task may be to determine a malodorous state based at least partially on the embedded output. The task may be to determine whether spoilage has occurred based at least partially on the embedded output. The task may include presenting a human-entered label for display, the human-entered label may be determined by its association with the embedded output in the embedded space. The human-entered label may describe the name of a particular food.

[0008] In some embodiments, a machine learning model can be trained in conjunction with a graph neural network, and training may involve jointly training the machine learning model and the graph neural network to produce a single combined output within the embedding space. The graph neural network may be trained to take a graph-based representation of a particular compound as input and output each embedding in the embedding space.

[0009] In some embodiments, a machine learning model may be trained by obtaining an example of compound training, which includes electrical signal training data and its respective training labels. The electrical signal training data and its respective training labels can describe a particular training compound. The machine learning model may be trained by processing the electrical signal training data with the machine learning model to generate compound embedding outputs, processing the compound embedding outputs with a classification model to determine the compound labels, evaluating a loss function that assesses the difference between the compound labels and their respective training labels, and tuning one or more parameters of the machine learning model, at least in part, based on the loss function.

[0010] In some embodiments, a machine learning model can be trained using supervised learning. Sensor data can describe at least one of voltage or current. The machine learning model may include a transformer model. In some embodiments, the operation may include storing an embedded output. Sensor data can describe the amplitude of one or both of the voltages or currents of one or more electrical signals. Processing the sensor data with a machine learning model using one or more processors to generate an embedded output in the embedding space may include compressing the sensor data into a fixed-length vector representation.

[0011] Another exemplary aspect of this disclosure relates to a computer implementation method. This method may include acquiring sensor data from one or more sensors by a computing system comprising one or more processors. In some embodiments, the sensor data may describe electrical signals generated due to the presence of one or more compounds in the environment. The method may include processing the sensor data using a machine learning model by the computing system to generate an embedded output in an embedding space. The machine learning model can be trained to receive and process data describing electrical signals and generate embeddings in the embedding space. The method may include determining one or more labels associated with the embedded output in the embedding space by the computing system, and presenting one or more labels for display by the computing system.

[0012] Another exemplary aspect of this disclosure relates to one or more non-temporary computer-readable media that, when executed by one or more processors, collectively store instructions causing a computing system to perform an action. The action may include acquiring sensor data using one or more sensors. In some embodiments, the sensor data may describe electrical signals generated due to the presence of one or more compounds in the environment. The method may include processing the sensor data using a machine-trained model to generate an embedded output in the embedding space. The machine-trained model can be trained to receive and process data describing electrical signals and generate embeddings in the embedding space. The action may include acquiring a plurality of stored sensory characteristic datasets, the plurality of stored sensory characteristic datasets may include stored embeddings in the embedding space paired with each sensory characteristic dataset associated with each stored embedding. The action may include determining one or more sensory characteristics based on the embedded output in the embedding space and the plurality of stored sensory characteristic datasets, and presenting one or more sensory characteristics for display.

[0013] Other aspects of this disclosure cover a variety of systems, apparatus, non-temporary computer-readable media, user interfaces, and electronic devices.

[0014] These and other features, aspects and advantages of the various embodiments of this disclosure will be better understood by referring to the following description and the appended claims. The appended drawings incorporated into this specification and forming part of this specification illustrate exemplary embodiments of this disclosure and, together with this description, serve to illustrate the relevant principles. The present invention provides, for example, the following items: (Item 1) A computing system, A sensor configured to generate an electrical signal indicating the presence of one or more compounds in the environment. A machine learning model that receives and processes the aforementioned electrical signals and is trained to generate embeddings in the embedding space, The machine learning model is trained using a training dataset that includes multiple training examples, each training example including a ground truth characteristic label applied to a set of electrical signals generated by one or more test sensors when exposed to one or more training compounds, and each ground truth characteristic label describes the properties of the one or more training compounds, the machine learning model, One or more processors, and The system includes one or more non-temporary computer-readable media that store together instructions causing the computing system to perform an operation when executed by the one or more processors, and the operation is The sensor generates sensor data indicating the presence of a specific compound in the environment, and The one or more processors process the sensor data using the machine learning model to generate an embedded output in the embedded space. The computing system including the said. (Item 2) A computing system as described in any of the preceding items, further comprising performing a task based on the aforementioned embedded output. (Item 3) A computing system according to any of the preceding items, wherein the task includes providing a prediction of sensory characteristics based on the embedded output. (Item 4) A computer system according to any of the preceding items, wherein the task includes providing an olfactory characteristic prediction based on the embedded output. (Item 5) The computing system described in any of the preceding items, wherein the task is to identify a disease state based at least in part on the embedded output. (Item 6) The computing system according to any of the preceding items, wherein the task is to determine the state of malodor based at least in part on the embedded output. (Item 7) The computing system described in any of the preceding items, wherein the task is to determine whether corruption has occurred based at least partially on the embedded output. (Item 8) The computing system described in any of the preceding items, wherein the task includes presenting a human-entered label for display, the human-entered label being determined by its association with the embedded output in the embedded space. (Item 9) The computing system described in item 8, wherein the labels entered by the aforementioned human describe the names of specific foods. (Item 10) The computing system described in any of the preceding items, wherein the machine learning model is trained in conjunction with a graph neural network, and the training comprises training the machine learning model and the graph neural network together to produce a single combined output within the embedding space. (Item 11) The computing system according to item 10, wherein the graph neural network is trained to take a graph-based representation of the particular compound as input and to output each of the embeddings in the embedding space. (Item 12) The aforementioned machine learning model, Obtaining an example of compound training including electrical signal training data and respective training labels, wherein the electrical signal training data and respective training labels describe a specific training compound, The process involves using the aforementioned machine learning model to process the electrical signal training data and generate a compound embedding output. processing an embedding output of said compound using a classification model to determine a compound label, evaluating a loss function that evaluates a difference between said compound label and said respective training labels, and adjusting one or more parameters of said machine-learned model based at least in part on said loss function, the computing system according to any one of the preceding items. (Item 13) the computing system according to any one of the preceding items, wherein said machine-learned model is trained by supervised learning. (Item 14) the computing system according to any one of the preceding items, wherein said sensor data describes at least one of voltage or current. (Item 15) the computing system according to any one of the preceding items, wherein said machine-learned model comprises a transformer model. (Item 16) the computing system according to any one of the preceding items, further comprising storing said embedding output. (Item 17) the computing system according to any one of the preceding items, wherein said sensor data describes the amplitude of one or both of voltage or current of one or more electrical signals. (Item 18) the computing system according to any one of the preceding items, wherein processing said sensor data with said machine-learned model by said one or more processors to generate said embedding output in said embedding space comprises compressing said sensor data into a fixed-length vector representation. (Item 19) A computer-implemented method comprising: The acquisition of sensor data by a computing system comprising one or more processors, wherein the sensor data describes an electrical signal generated by the presence of one or more compounds in the environment. The computing system processes the sensor data using a machine learning model and generates an embedded output in the embedded space, wherein the machine learning model is trained to receive and process data describing electrical signals and generate an embedded in the embedded space, The computing system determines one or more labels associated with the embedded output of the embedded space, and The computing system presents the one or more labels for display. The method, including the method described above. (Item 20) One or more non-temporary computer-readable media that store a set of instructions causing a computing system to perform an action when executed by one or more processors, wherein the action is: Acquiring sensor data with one or more sensors, wherein the sensor data describes an electrical signal generated by the presence of one or more compounds in the environment, The process involves processing the sensor data using a machine learning model and generating an embedded output in the embedded space, wherein the machine learning model is trained to receive and process data describing electrical signals and generate an embedded in the embedded space, and the generation process involves processing the sensor data using a machine learning model and generating an embedded output in the embedded space. Acquiring a plurality of stored sensory characteristic datasets, wherein each of the plurality of stored sensory characteristic datasets includes a stored embedding in the embedding space, paired with each sensory characteristic dataset associated with each of the stored embeddings. Determining one or more sensory characteristics based on the embedded output of the embedded space and the plurality of stored sensory characteristic datasets, and To present one or more of the aforementioned sensory characteristics for display purposes, One or more non-temporary computer-readable media, including [the specified text].

[0015] A detailed description of embodiments with reference to the accompanying drawings, intended for those skilled in the art, is provided herein. [Brief explanation of the drawing]

[0016] [Figure 1A] A block diagram of an exemplary computing system that performs sensor data processing according to exemplary embodiments of this disclosure is shown. [Figure 1B] A block diagram of an exemplary computing device that performs sensor data processing according to exemplary embodiments of the present disclosure is shown. [Figure 1C] A block diagram of an exemplary computing device that performs sensor processing according to an exemplary embodiment of the present disclosure is shown. [Figure 2] A block diagram of an exemplary classification process according to an exemplary embodiment of the present disclosure is shown. [Figure 3] A block diagram of an exemplary electrochemical sensor system according to an exemplary embodiment of the present disclosure is shown. [Figure 4] A block diagram of an exemplary training process according to an exemplary embodiment of this disclosure is shown. [Figure 5] A block diagram of exemplary sensor data machine learning model processing according to exemplary embodiments of this disclosure is shown. [Figure 6] A flowchart illustrating an exemplary method for performing sensor data processing according to an exemplary embodiment of this disclosure is shown. [Figure 7] A flowchart illustrating an exemplary method for performing sensor data processing according to an exemplary embodiment of this disclosure is shown. [Figure 8]A flowchart illustrating an exemplary method for performing machine learning model training according to exemplary embodiments of this disclosure is shown. [Figure 9] A block diagram of an exemplary training process according to an exemplary embodiment of this disclosure is shown. [Modes for carrying out the invention]

[0017] The repeated reference numbers across multiple diagrams are intended to identify the same features in various implementations.

[0018] overview Generally, this disclosure relates to the processing of sensor data describing the presence of chemical molecules. Systems and methods can be used for electrical signal processing to enable the interpretation of sensor data obtained from electrochemical sensor devices. Systems and methods disclosed herein can process sensor data using a trained machine learning model and subsequently generate embedded outputs in an embedding space that can be used to perform various tasks. Ground truth datasets can be used to train the machine learning model, and existing databases of chemical molecule property data can also be utilized.

[0019] More specifically, in some embodiments, the systems disclosed herein may include a sensor configured to generate an electrical signal. The electrical signal may indicate the presence of one or more compounds in the environment, and a machine-trained model may be trained to receive and process the electrical signal to generate an embedding in the embedding space. The machine-trained model may be trained using a training dataset that includes multiple training examples. The training examples may include ground truth characteristic labels applied to each set of electrical signals generated by the sensor when exposed to one or more training compounds. The ground truth characteristic labels may describe the characteristics of one or more training compounds. Furthermore, the system may include one or more processors and one or more non-temporary computer-readable media that collectively store instructions causing a handheld remote control device to perform an action when executed by one or more processors. Including these components allows the sensor to generate sensor data based on the electrical signal, which can then be processed by the machine-trained model to generate an embedding output in the embedding space. More specifically, the systems and methods disclosed herein can be used to generate sensor data that describes the electrical signal generated when the chemical characteristics of a sensor react with compounds in the environment. Sensor data can then be processed by a machine learning model to generate embedded outputs in the embedded space. In some embodiments, the embedded space can be populated with embeddeds generated based on electrical signals and embeddeds generated based on graphical representations of compounds. Furthermore, in some embodiments, the embedded space can be populated with embedded labels describing the names or properties of chemical mixtures, which may be generated based on human input or automated predictions.

[0020] In some embodiments, the system and method may further include performing tasks based on the embedded output. Tasks may include providing classification output, determining characteristic predictions, presenting alerts, and / or storing the embedded output. For example, the embedded output may be processed to determine one or more characteristic predictions, which may then be presented for display to the user. Characteristic predictions may be sensory characteristic predictions, such as olfactory characteristic predictions or volatility predictions, and determining these may lead to the presentation of a hazardous chemical alarm.

[0021] In some embodiments, a machine learning model can be trained by obtaining multiple training examples, each containing an electrical signal dataset and its respective training label. The electrical signal training dataset and its respective training label can describe a specific compound. The electrical signals can be processed to generate embedding outputs. These embedding outputs can then be processed by a classification model to determine the respective compound label for each electrical signal dataset. The resulting labels can be compared to ground truth labels to determine if the parameters of the machine learning model need to be adjusted. Furthermore, in some embodiments, the machine learning model may be trained in conjunction with a graph neural network (GNN) model to generate embeddings using graph representations or electrical signals, which can then be used for classification tasks. In some embodiments, supervised learning can be included in the training.

[0022] The trained machine learning models can then be used for a variety of tasks, such as predicting sample characteristics based on electrical signals, determining whether crops are diseased, identifying food spoilage, diagnosing diseases, and detecting the presence or absence of foul odors. The machine learning models can be stored locally on a computing device as part of an electrochemical sensor device, or they can be stored and accessed as part of a larger computing system. The systems and processes can be used for a variety of applications, whether personal, commercial, or industrial.

[0023] An electrochemical sensor may include one or more sensors and optionally one or more processors. The device can use one or more sensors to acquire sensor data describing the environment. The sensor data may describe compounds in the environment. In some embodiments, the sensor data can be processed to determine the composition of a mixture. The sensor data can be processed with a machine learning model to determine the mixture. The determination of a mixture may involve processing the sensor data to generate embeddings, which are then processed with a classification model to determine the composition of the mixture. In some embodiments, the determination process may utilize labeled embedding spaces generated using labeled embeddings. The determined mixture may be determined based on one or more determined mixture labels in the labeled embedding spaces.

[0024] Calibrating an electrochemical sensor device to determine a mixture or property may involve obtaining multiple mixture datasets. Each mixture dataset can describe one or more sensory properties of each mixture. For each of the multiple mixtures, one or more mixture labels can be obtained. The multiple mixture datasets can be processed with a machine learning model to generate multiple mixture embeddings. Each mixture embedding can be associated with its respective mixture dataset. The multiple embeddings can then be paired with their respective mixture labels. A labeled embedding space can be generated using these labeled embeddings.

[0025] In some embodiments, the mixture label may be a human-entered label. In some embodiments, the system may collect human-labeled, accurate sensor data (e.g., human-labeled odor data) for calibration. The calibrated electrochemical sensor device can then detect chemicals consisting of a mixture of molecules, each with potentially different concentrations. In some embodiments, one or more sensors may include nasal electronic sensors capable of generating sensor data. The sensor data may describe electronic signals. One or more sensors may include, but are not limited to, carbon nanotubes, DNA-bound carbon nanotubes, carbon black polymers, photosensitive chemical sensors, sensors constructed from silicon-bonded biosensors, olfactory neurons cultured from stem cells or taken from living organisms, olfactory receptors, and / or metal oxide sensors. The resulting sensor data may be raw data, including voltage data or current data.

[0026] In some embodiments, experiments in which both human labels and electronic signals can be collected from the same or very similar samples can be used for calibration. In some embodiments, the machine-learned model can be trained using ground truth training data that includes multiple sensory datasets and multiple mixture labels. The machine-learned model may include one or more transformer models and / or one or more GNN embedding models.

[0027] Furthermore, the calibration of an electrochemical sensor device may include mapping human labels to an embedding space (e.g., an odor embedding space). A trained GNN can be used for the mapping. In this case, the use of the device may include mapping the obtained electrical signals to the embedding space. The mapped locations (i.e., values ​​in the embedding space) can be used to automatically recognize odors or other sensory characteristics with human labels such as "cinnamon," "cucumber," "apple," and "feces." The mapping of electrical signals can be performed using a deep neural network, with a GNN trained on the electro-nasal signals. In some embodiments, the embeddings can be configured similarly to RGB numbering. In some embodiments, processing the sensor data and embedding space may include processing the sensor data with a machine learning model to generate embeddings, mapping the embeddings to the embedding space, and determining matching labels based on the location of the embeddings associated with one or more mixture labels.

[0028] The accuracy of predicting human scents can be evaluated using electronic sensor signals. Low accuracy for specific human scents, such as "cinnamon," may indicate that the sensor cannot accurately detect that scent. High accuracy for specific scents may indicate that the sensor can accurately detect that scent.

[0029] In some embodiments, an electrochemical sensor can consist of a number of distinct sensing elements, similar to how a camera can sense both red and green colors. Using this system, which collects both human-labeled data and electronic signal data, the system can evaluate whether a new sensing element (assuming the camera can now sense blue) improves the ability to cover a human-perceivable spatial range of odors or to recognize specific odor labels.

[0030] Instead of recognizing odor labels defined by humans, the system can instead define labels as the presence or absence of diseased humans, animals, or plants that emit a characteristic odor.

[0031] In some embodiments, the systems and methods disclosed herein may be implemented to identify food or specific flavors based on collected sensor data. For example, a glass of orange juice can be placed under a sensor to generate sensor data describing exposure to one or more chemicals. The sensor data can be processed by a machine learning model to generate an embedding output in an embedding space. The embedding output can then be used to determine food labels and / or flavor labels. For example, the embedding output may be determined to be most similar to the embedding combined with an orange label or an orange juice label. In some embodiments, the embedding output can be analyzed to determine that the sensed chemicals indicate a citrus flavor. Determining food type and flavor may involve classification models, threshold determination, and / or analysis of labeled embedding spaces or maps.

[0032] Another exemplary use of the systems and methods disclosed herein may include the activation of diagnostic sensors for human, animal, or plant diagnosis. The presence of certain chemicals may indicate a particular disease state. For example, compounds found in human exhalation may provide valuable information about the presence or stage of certain diseases or illnesses (such as gastroesophageal reflux disease, periodontitis, periodontal disease, diabetes, liver or kidney disease, etc.). Thus, in some embodiments, sensor data can describe exposure to chemicals exhaled or taken as samples from patients. Sensor data can be processed by a machine learning model to generate embedding outputs. The embedding outputs can be compared to embeddings indicating a perceived disease state, or processed by a classification head trained for diagnosis to determine whether chemicals indicating a disease state are present. The output of the classification head may include the probability of each of one or more disease states being present.

[0033] Electrochemical sensor devices can be implemented in cooking appliances such as stoves and exhaust hoods to assist with cooking and provide alerts regarding the cooking process. In some embodiments, the electrochemical sensor device can be implemented to provide an alert if chemicals indicating burnt food are present. For example, an embedded output can be input to a classification head, which processes the embedded output to determine the probability of burnt food being present. If the probability exceeds a threshold probability, an alert may be activated.

[0034] Furthermore, in some embodiments, electrochemical sensor devices equipped with trained machine learning models can be implemented on agricultural equipment such as ground vehicles or low-altitude UAVs to detect the presence of diseased crops or whether plants are ripe for harvest. For example, the embedded output can be input to a classification head, which processes the embedded output to determine the probability that the plants are ripe for harvest.

[0035] In some embodiments, the systems and methods disclosed herein may be used to control machinery and / or to issue alerts. The systems and methods can be used to control manufacturing machinery to provide a safer working environment or to modify the composition of a mixture to achieve a desired yield. Furthermore, in some embodiments, real-time sensor data can be generated and processed to produce embedded outputs that can be classified to determine whether an alert (e.g., an alert indicating a hazardous condition, food spoilage, disease state, or foul odor) should be issued. For example, in some embodiments, the determined classification may include characteristic predictions, such as olfactory characteristic predictions for the scent of a vehicle used in a transport service. The classification can then be processed to determine when a new scent product should be placed in the transport device and / or whether the transport device should undergo a cleaning routine. The determination that a foul odor is present may then be sent as an alert to the user's computing device or used to set up an automatic purchase. In another example, a transport device (e.g., an autonomous vehicle) may be automatically recalled to a facility to undergo a cleaning routine. In yet another example, an alert may be issued if characteristic predictions generated by a machine learning model indicate the presence of a dangerous environment within the space for animals or people. For example, if a safety deficiency is predicted based on chemicals detected in a building, an audio alert can be sounded in the building. For instance, embedded output can be input to a classification head, which can process the output to determine the probability that the environment contains hazardous chemicals. If the probability exceeds a threshold, an alert can be issued and / or an alarm can be triggered.

[0036] In some embodiments, the system can take in sensor data to be input into an embedding model and a classification model to generate predictions of environmental characteristics. For example, the system can utilize one or more sensors to take in data related to the presence and / or concentration of molecules in the environment. The system can process the sensor data to generate input data for the embedding model and process the classification model to generate predictions of environmental characteristics. This may include one or more predictions about the smell of the environment or other characteristics of the environment. If the predictions include a particular unpleasant odor, the system can send an alert to the user's computing device to complete a cleaning service. In some embodiments, once the system detects an unpleasant odor, it can bypass the alert and send a booking request to the cleaning service.

[0037] Another exemplary embodiment may include background processing and / or active monitoring for safety measures. For example, the system may actively generate and process sensor data acquired by sensors in a manufacturing plant to ensure that the manufacturer is aware of any hazards. In some embodiments, the sensor data may be generated intermittently or intermittently and processed by embedded and classification models to determine characteristic predictions. Characteristic predictions may include whether a chemical in the environment is flammable, toxic, unstable, or hazardous in any way. For example, a characteristic prediction may include a probability score for each of several environmentally hazardous conditions that exist. If a chemical sensed in the environment is determined to be hazardous in any way, for example, if the probability score for any one or more environmentally hazardous conditions exceeds their respective thresholds, an alert may be sent. Alternatively and / or additionally, the system may control one or more machines to stop and / or suppress processing in order to protect against any potential current or future hazards.

[0038] The systems and methods can be applied to other manufacturing, industrial, or commercial systems to generate automated alerts or actions based on characteristic predictions. These applications may include identifying sensed chemicals, determining the characteristics of sensed chemicals, identifying diseases, identifying food spoilage, and determining crop problems.

[0039] In some embodiments, the systems and methods disclosed herein can utilize a chemical mixture property prediction database to classify the embedded outputs. The database may be generated by generating theoretical chemical mixture property predictions using an embedded model and a prediction model for determining the predicted properties.

[0040] For example, the system and method may include obtaining molecular data for one or more molecules and mixture data relating to a mixture of one or more molecules. The molecular data may include individual molecular data for each of the multiple molecules that make up the mixture. In some embodiments, the mixture data may include data relating to the concentration of each molecule in the mixture, along with the overall composition of the mixture. The mixture data may describe the chemical composition of the mixture. The molecular data may be processed in an embedding model to generate multiple embeddings. Each molecular data for each molecule may be processed in the embedding model to generate a separate embedding for each molecule in the mixture. In some embodiments, the embeddings may include data describing the individual molecular properties of the embedding data. In some embodiments, the embeddings may be numerical vectors. In some cases, the embeddings may represent graphs or descriptions of molecular properties. The embedding and mixture data may be processed by a predictive model to generate one or more characteristic predictions. One or more characteristic predictions may be at least partially based on one or more embedding and mixture data. Characteristic predictions may include various predictions regarding the taste, smell, color, etc., of the mixture. In some embodiments, the system and method may include storing one or more characteristic predictions. In some embodiments, one or both of the models may include a machine learning model.

[0041] Next, labeled embeddings can be generated in the embedding space by pairing the embeddings with their respective characteristic predictions as labeled sets. A machine learning model can then be trained to output embedding outputs that can be subsequently compared with labels in the embedding space for classification tasks such as determining the characteristics of a sensed compound or identifying a chemical mixture sensed by a sensor.

[0042] The systems and methods of this disclosure offer numerous technical effects and advantages. For example, the systems and methods can provide devices and processes that enable the understanding and interpretation of electrical signals, which can lead to efficient and accurate identification processes. The systems and methods can further be used to identify food spoilage using electrical sensors, or to identify disease states in plants, animals, or humans. Furthermore, the systems and methods can enable automated processes for identifying compounds based on electrical signal data generated by electrochemical sensors.

[0043] Another technical advantage of the systems and methods of this disclosure is the ability to utilize odor embedding spaces for the classification of electrical signals. While manually training a model to identify all known mixtures or properties can be cumbersome, using a generated odor embedding space allows for the presentation of easily accessible data without having to start training from scratch.

[0044] Another example of the technical effects and benefits concerns improved computational efficiency and enhanced computing system capabilities. For example, certain existing systems are trained to identify the presence of a single compound or a small number of compounds. Training each compound individually can be time-consuming, and computations can be inefficient when the system only tests whether a compound is present or absent. In contrast, by training a machine learning model to generate embedded outputs in the embedding space, the system can leverage the embedding properties to efficiently determine a compound or chemical property. Thus, the proposed system and method can save computational resources such as processor usage, memory usage, and / or network bandwidth.

[0045] Exemplary embodiments of this disclosure will be described in further detail here with reference to the drawings.

[0046] Exemplary devices and systems Figure 1A shows a block diagram of an exemplary computing system 100 that performs electrical signal processing according to an exemplary embodiment of the present disclosure. The system 100 includes a user computing device 102, a server computing system 130, and a training computing system 150, all of which are communicably coupled via a network 180.

[0047] The user computing device 102 may be any type of computing device, such as a personal computing device (e.g., a laptop or desktop), a mobile computing device (e.g., a smartphone or tablet), a game console or controller, a wearable computing device, an embedded computing device, or any other type of computing device.

[0048] The user computing device 102 includes one or more processors 112 and memory 114. The one or more processors 112 can be any suitable processing device (e.g., a processor core, microprocessor, ASIC, FPGA, controller, microcontroller, etc.) and can be one processor or multiple processors connected in an operable manner. The memory 114 can include one or more non-temporary computer-readable storage media such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, and combinations thereof. The memory 114 can store data 116 and instructions 118 that are executed by the processors 112 to cause the user computing device 102 to perform operations.

[0049] In some embodiments, the user computing device 102 may store or include one or more electrical signal processing models 120. For example, the electrical signal processing models 120 may be, or otherwise include, various machine learning models such as neural networks (e.g., deep neural networks) or other types of machine learning models including nonlinear and / or linear models. Neural networks may include feedforward neural networks, recurrent neural networks (e.g., long-term short-term memory recurrent neural networks), convolutional neural networks, or other forms of neural networks. Exemplary electrical signal processing models 120 are described with reference to Figures 4, 5, and 9.

[0050] In some embodiments, one or more electrical signal processing models 120 may be received from a server computing system 130 via a network 180, stored in a user computing device memory 114, and then used or otherwise implemented by one or more processors 112. In some embodiments, the user computing device 102 may implement multiple parallel instances of a single electrical signal processing model 120 (for example, to perform parallel electrical signal processing across multiple instances of different compounds being sensed).

[0051] More specifically, an electrical signal processing model could be a machine learning model trained to receive sensor data describing electrical signals representing a compound, process the sensor data, and output an embedded output to an embedding space. This embedded output can then be used to perform various tasks. For example, the embedded output can be processed by a classification model to determine the compound's molecules and concentrations, or its properties. The results can then be presented to the user.

[0052] In addition, or alternatively, one or more electrical signal processing models 140 may be included in, or stored and implemented by, a server computing system 130 that communicates with a user computing device 102 according to a client-server relationship. For example, an electrical signal processing model 140 may be implemented by the server computing system 140 as part of a web service (e.g., an electrochemical sensor service). Thus, one or more models 120 may be stored and implemented in the user computing device 102, and / or one or more models 140 may be stored and implemented in the server computing system 130.

[0053] The user computing device 102 may also include one or more user input components 122 that receive user input. For example, a user input component 122 could be a touch-sensitive component (e.g., a touch-sensitive display screen or touchpad) that senses the touch of a user input object (e.g., a finger or stylus). The touch sensor component functions to implement a virtual keyboard. Other exemplary user input components include a microphone, a conventional keyboard, or other means by which the user can perform user input.

[0054] The server computing system 130 includes one or more processors 132 and memory 134. The one or more processors 132 can be any suitable processing device (e.g., a processor core, microprocessor, ASIC, FPGA, controller, microcontroller, etc.) and can be one processor or multiple processors connected in an operable manner. The memory 134 can include one or more non-temporary computer-readable storage media such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, and combinations thereof. The memory 134 can store data 136 and instructions 138 that are executed by the processors 132 to cause the server computing system 130 to perform operations.

[0055] In some embodiments, the server computing system 130 includes or is implemented by one or more server computing devices. If the server computing system 130 includes multiple server computing devices, such server computing devices can operate according to a sequential computing architecture, a parallel computing architecture, or a combination thereof.

[0056] As described above, the server computing system 130 may store or otherwise include one or more machine learning-trained electrical signal processing models 140. For example, the models 140 may be or may include various machine learning-trained models. Exemplary machine learning-trained models include neural networks or other multilayer nonlinear models. Exemplary neural networks include feedforward neural networks, deep neural networks, recurrent neural networks, and convolutional neural networks. Exemplary models 140 are described with reference to Figures 4, 5, and 9.

[0057] The user computing device 102 and / or the server computing system 130 can train models 120 and / or 140 through interaction with a training computing system 150 which is communicatively coupled via a network 180. The training computing system 150 may be separate from the server computing system 130 or may be part of the server computing system 130.

[0058] The training computing system 150 includes one or more processors 152 and memory 154. The one or more processors 152 can be any suitable processing device (e.g., a processor core, microprocessor, ASIC, FPGA, controller, microcontroller, etc.) and can be one processor or multiple processors connected in an operable manner. The memory 154 can include one or more non-temporary computer-readable storage media such as RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. The memory 154 can store data 156 and instructions 158 executed by the processors 152 to cause the training computing system 150 to perform operations. In some embodiments, the training computing system 150 includes or is implemented by one or more server computing devices.

[0059] The training computing system 150 may include a model trainer 160 that trains machine-learned models 120 and / or 140 stored in the user computing device 102 and / or server computing system 130 using various training or learning techniques, such as backpropagation of errors. For example, a loss function may be backpropagated through the model(s) to update one or more parameters of the model(s) (e.g., based on the gradient of the loss function). Various loss functions can be used, such as mean squared error, likelihood loss, cross-entropy loss, hinge loss, and / or various other loss functions. Using gradient descent, training can be repeated many times to repeatedly update parameters.

[0060] In some embodiments, performing error backpropagation may include performing censored diachronic backpropagation. The model trainer 160 can perform many generalization techniques (e.g., weight decay, dropout, etc.) to improve the generalization ability of the model during training.

[0061] In particular, the model trainer 160 can train the electrical signal processing models 120 and / or 140 based on a set of training data 162. The training data 162 may include, for example, a paired set of data, each paired set including electrical signal training data and a ground truth training label for each electrical signal training data.

[0062] In some embodiments, if the user consents, training examples may be presented by the user computing device 102. Thus, in such embodiments, the model 120 provided to the user computing device 102 can be trained by the training computing system 150 based on user-specific data received from the user computing device 102. In some cases, this process may be referred to as model personalization.

[0063] The model trainer 160 includes computer logic used to produce a desired function. The model trainer 160 can be implemented with hardware, firmware, and / or software that controls a general-purpose processor. For example, in some embodiments, the model trainer 160 includes a program file that is stored in a storage device, loaded into memory, and executed by one or more processors. In other embodiments, the model trainer 160 includes one or more sets of computer executable instructions stored in a tangible computer-readable storage medium such as a RAM hard disk, optical media, or magnetic media.

[0064] Network 180 can be any type of communication network, such as a local area network (e.g., an intranet), a wide area network (e.g., the Internet), or a combination thereof, and may include any number of wired or wireless links. In general, communication on Network 180 can be carried out over any type of wired and / or wireless connection using a wide variety of communication protocols (e.g., TCP / IP, HTTP, SMTP, FTP), encodings or formats (e.g., HTML, XML), and / or protection schemes (VPN, Secure HTTP, SSL, etc.).

[0065] Figure 1A shows an example of a computing system that can be used to implement the present disclosure. Other computing systems can be used in a similar manner. For example, in some embodiments, the user computing device 102 may include a model trainer 160 and a training dataset 162. In such embodiments, the model 120 can be trained and used locally on the user computing device 102. In some such embodiments, the user computing device 102 may implement the model trainer 160 to personalize the model 120 based on user-specific data.

[0066] Figure 1B shows a block diagram of an exemplary computing device 10 that operates according to an exemplary embodiment of the present disclosure. The computing device 10 may be a user computing device or a server computing device.

[0067] Computing device 10 contains numerous applications (e.g., applications 1 through N). Each application contains its own machine learning library and pre-trained model(s). For example, each application may contain a pre-trained model. Exemplary applications include text messaging applications, email applications, dictation applications, virtual keyboard applications, and browser applications.

[0068] As shown in Figure 1B, each application can communicate with many other components of the computing device, such as one or more sensors, a context manager, a device state component, and / or additional components. In some embodiments, each application can communicate with each device component using an API (e.g., a public API). In some embodiments, the API used by each application is specific to that application.

[0069] Figure 1C shows a block diagram of an exemplary computing device 50 that operates according to an exemplary embodiment of the present disclosure. The computing device 50 may be a user computing device or a server computing device.

[0070] The computing device 50 contains numerous applications (applications 1 through N, etc.). Each application communicates with a central intelligence layer. Exemplary applications include text messaging applications, email applications, dictation applications, virtual keyboard applications, and browser applications. In some embodiments, each application can communicate with the central intelligence layer (and the models(s) stored therein) using an API (such as a common API across all applications).

[0071] The central intelligence layer contains numerous machine learning models. For example, as shown in Figure 1C, each machine learning model (e.g., Model) can be provided for each application and managed by the central intelligence layer. In other embodiments, two or more applications can share a single machine learning model. For example, in some embodiments, the central intelligence layer can provide a single model (e.g., Single Model) for all applications. In some embodiments, the central intelligence layer is contained within or implemented by the operating system of the computing device 50.

[0072] The central intelligence layer can communicate with the central device data layer. The central device data layer may be a centralized repository of data for the computing device 50. As shown in Figure 1C, the central device data layer can communicate with many other components of the computing device, such as one or more sensors, a context manager, a device state component, and / or additional components. In some embodiments, the central device data layer can communicate with each device component using an API (e.g., a private API).

[0073] Exemplary model placement Figure 2 shows a block diagram of an exemplary bipedal classification system 200 according to an exemplary embodiment of the present disclosure. In some embodiments, the bipedal classification system 200 receives either a graph representation 210 of a compound or electrical signal data 220 describing a compound, and is trained to present output data 230 that classifies the input data as being associated with a particular compound or a particular property as a result of receiving the input data 210 and 220. Thus, in some embodiments, the bipedal classification system 200 may include a graph neural network 212 capable of processing the graph representation 210 and a machine learning model 222 capable of processing the electrical signal data 220.

[0074] In particular, Figure 2 shows a system 200 that can provide classification by processing either sensor data or graph representation data. The illustrated system 200 includes a first foot for processing graph representations of one or more molecules 210 and a second foot 220 for processing electrical signal data or sensor data of one or more molecules. However, in some embodiments, both graph representations 210 and sensor data 220 can be processed in a single model architecture.

[0075] Processing the graph representation 210 may include processing the data describing the graph representation 210 using a graph neural network (GNN) model 212 to generate embeddings 214. The embeddings may be based at least partially on molecular concentrations. The embeddings 214 may be embeddings in the embedding space.

[0076] Processing of the electrical signal data 220 may include processing the electrical signal data 220 using a machine learning model 222 to generate an ML output 224. In some embodiments, the electrical signal data 220 may be acquired from or generated using one or more sensors. One or more sensors may include electrochemical sensors. Furthermore, in some embodiments, the electrical signal data 220 may include sensor data describing one or more electrical signals generated in response to exposure to a compound. The machine learning model 222 may include one or more embedding models and / or one or more transformer models. Furthermore, the ML output 224 may be an embedding output in the embedding space.

[0077] In some embodiments, the GNN model 212 and the machine-trained model 22 can be trained to provide embeddings 214 and embedding outputs 224 in the same embedding space. Furthermore, in some embodiments, the GNN model 212 and the machine-trained model 222 may be a single shared model. The two models may be part of the same model architecture.

[0078] Next, the embeddings 214 and ML output 224 can be processed by a classification model to determine the classification 230. The classification 230 can be at least partially based on a set of human-input labels. In some embodiments, the classification 230 can be at least partially based on predictive property labels in the embedding space. The predictive property labels can be at least partially based on a chemical mixture property prediction system that utilizes an embedding model and a predictive model to determine the theoretical property prediction of a mixture.

[0079] Figure 3 shows a block diagram of an exemplary electrochemical sensor device system 300 according to an exemplary embodiment of the present disclosure. In some embodiments, the electrochemical sensor device system 300 may include a sensor computing system 310 comprising a machine learning model 312, one or more sensors 314, a user interface 316, a processor 318, memory 320, and a GNN embedded model 330.

[0080] In particular, the sensor computing system 310 may include an electrochemical sensor device that includes one or more sensors 314 for sensing exposure to a compound. The sensors 314 can be configured to generate sensor data that describes electrical signals obtained in response to exposure to one or more molecules.

[0081] Furthermore, the sensor computing system 310 may include a machine learning model 312 for processing sensor data and generating embedded outputs in the embedding space. The sensor computing system may further include an embedding model 330 for processing graph representations and / or for training the machine learning model 312 in conjunction with a graph neural network embedding model 330.

[0082] In some embodiments, the sensor computing system may include one or more memory components 320 for storing embedded spatial data 322, electrical signal data 324, labeled dataset 326, other data, and instructions for performing one or more actions or functions. In particular, memory 320 may store embedded spatial data 322 generated using a database of embedded label pairs. For example, the embedded spatial data 322 may include a set of multiple pairs, each containing an embedding generated based on a graph representation or sensor data, and a label describing a chemical mixture or characteristic prediction. The embedded spatial data 322 can assist in classification tasks, such as determining the compound to which the sensor has been exposed.

[0083] The memory component may also store historical electrical signal data 324 and labeled data 326. Historical electrical signal data 324 may be stored for training, classification tasks, and / or to maintain a data log of historical acquisition data. For example, a set of electrical signal data 324 may not reach the threshold classification score of any of the stored labels or classes, and therefore may be stored as a new classification label or class. However, in some embodiments, the electrical signal data 324 may include deviations from the training data that match the classification threshold. The sensor computing system may record historical electrical signal data 324 or historical sensor data to determine recurring deviation trends or errors that may indicate the need for sensor calibration or parameter adjustment.

[0084] Alternatively and / or additionally, the memory component 320 can store a labeled dataset 326 in place of, or in combination with, the embedded spatial data 322. The labeled dataset 326 can be used for classification tasks or for training a machine learning model 312. In some embodiments, the sensor computing system 310 can actively take in human-entered labels to improve the accuracy of classification tasks or for future training.

[0085] The sensor computing system may include a user interface 316 for receiving user input and providing notifications and feedback to the user. For example, in some embodiments, the sensor computing system 310 may include a display on or attached to the electrochemical sensor that can display a user interface providing notifications about embedded values, sensor data classifications, etc. In some embodiments, the electrochemical sensor may include a touchscreen display for receiving user input to assist in the use of the electrochemical sensor.

[0086] The sensor computing system 310 can communicate with one or more other computing systems via the network 350. For example, the sensor computing system 310 can communicate with a server computing system 360 via the network 350. The server computing system 360 may include a machine learning model 362, a graph neural network embedded model 364, stored data 366, and one or more processors 368. In some embodiments, the server computing system 360 may receive sensor data or labeled data 326 from the sensor computing system to assist with retraining or diagnostic tasks of the machine learning model. In some embodiments, the stored data 366 of the server computing system 360 may include a labeled embedded database accessible by the sensor computing system 310 via the network to assist with classification tasks and training. In some embodiments, the server computing system 360 may provide updated models to one or more sensor computing systems 310. Furthermore, in some embodiments, the sensor computing system 310 may utilize one or more processors 368 and the machine learning model 362 of the server computing system 360 to process sensor data generated by one or more sensors 314.

[0087] In some embodiments, the sensor computing system 370 can communicate with one or more other computing devices 370 to present notifications, process sensor data from other computing devices 370, or perform other computing tasks.

[0088] Figure 4 shows a block diagram of an exemplary system 400 for training a machine learning model according to an exemplary embodiment of the present disclosure. In some embodiments, the system 400 for training a machine learning model may include receiving an input dataset 404 describing compounds and training a machine learning model 410 to provide output data 416 describing predicted characteristic labels or chemical mixture labels as a result of receiving the input data 404. Thus, in some embodiments, the system 400 for training a machine learning model may include a classification model 414 capable of classifying the generated embeddings 412.

[0089] A machine learning model can be trained using ground truth labels. In some embodiments, the machine learning model may be an embedding model 410 trained to process sensor data 408 and produce a generated embedding output 412, which can then be used for a variety of other tasks.

[0090] In some embodiments, training of the embedded model 400 can begin with one or more training chemicals having human-labeled characteristics 402. One or more chemicals 404 can be exposed to one or more sensors 406 to generate sensor data describing the exposure to one or more chemicals 404. In some embodiments, the sensor data can describe electrical signals (e.g., voltage or current) generated by the electrochemical sensors.

[0091] The generated sensor data 408 may then be processed by an embedding model 410 to produce an embedding output 412. The embedding model 410 may include one or more transformer models. In some embodiments, the embedding model 410 may include a graph neural network model that can be trained to process both the graph representation and the sensor data 408. Furthermore, the generated embedding 412 may be an embedding output of the embedding space and may include a set of identifier values ​​similar to RGB values ​​for color representation.

[0092] Next, the generated embeddings 412 are processed by the classification head 414 to determine one or more matching predicted characteristic labels 416. The predicted characteristic labels 416 may include sensory characteristic labels such as smell, taste, or color. Then, the loss function 422 can be evaluated using the predicted characteristic labels 416 and the human-input characteristic labels 420. Next, one or more parameters of the machine-learned model 410 can be tuned by backpropagating the loss to learn / optimize the model parameters 418.

[0093] Process 400 can be iteratively completed over multiple training examples to train a machine learning model 410 to perform a classification task or generate an embedded output 412 that can be used to perform other tasks based on acquired sensor data 408.

[0094] Figure 5 shows a block diagram of an exemplary trained machine learning model system 500 according to an exemplary embodiment of the present disclosure. In some embodiments, the trained machine learning model system 500 receives a set of input data 504 describing one or more chemical substances and is trained to present output data 512 including generated embeddings as a result of receiving the input data 504. Thus, in some embodiments, the trained machine learning model system 500 may include a classification head 514 that can operate to determine a predicted characteristic label 516.

[0095] The trained machine learning model 510 can then be used for a variety of tasks, including characteristic prediction tasks.

[0096] For example, one or more chemical substances 502 can be exposed to one or more sensors 506 504 to generate sensor data 508. One or more sensors 506 may include one or more electrochemical sensors that can generate sensor data 508 describing electrical signal data observed while being exposed to one or more chemical substances 502. Furthermore, one or more chemical substances 502 can be exposed to one or more sensors 506 504 in a controlled environment (e.g., a laboratory space) or an uncontrolled environment (e.g., a car, an office, etc.).

[0097] The sensor data 508 may then be processed by a trained embedding model 510 to generate an embedding output 512. The embedding output 512 can be an embedding of the embedding space and may contain multiple values ​​that describe vector values.

[0098] In some embodiments, the embedding output 512 alone may be useful for clustering similar chemicals based on embeddings generated from sensor data of different chemicals 520. The embedding output 512 can also be used to better understand the properties of the embedding space and the different chemicals within it. Alternatively, and / or further, the embedding output alone may be used for a variety of tasks, which may include generating visualizations of the embedding space to provide a more intuitive depiction of the chemical properties space. The generated embedding output can then be used for further model training or a variety of other tasks.

[0099] Other applications of the embedded output 512 may include a classification task 518, which may include processing the embedded output 512 with a classification head 514 to determine one or more relevant predictive characteristic labels 516. The classification head 514 can be trained for characteristic prediction tasks, such as olfactory characteristic prediction, which can be used to determine when a car needs cleaning or when a foul odor is present.

[0100] Alternatively and / or additionally, the embedded output 512 can be processed by different heads trained for different tasks 522 to produce a predictive task output 524 that assists in the execution of task 524. In some embodiments, different heads 522 can be trained to classify whether the embedded output describes food spoilage, a disease state, or whether a chemical substance may have beneficial properties such as antifungal properties.

[0101] Figure 9 shows a block diagram of an exemplary system 900 for training a machine learning model according to an exemplary embodiment of the present disclosure. System 900 for training a machine learning model is similar to system 400 for training a machine learning model in Figure 4, except that it further includes training the system to handle graph representations.

[0102] In some embodiments, the machine learning models 910 and 926 can be trained using ground truth labels. In some embodiments, the machine learning models may be embedding models 910 and 926 trained to process data describing sensor data 908 and / or graph representations 924 to produce a generated embedding output 912, which can then be used for a variety of other tasks.

[0103] In some embodiments, training of the embedded model 900 can begin with one or more training chemicals having human-labeled characteristics 902. One or more chemicals 904 can be exposed to one or more sensors 906 to generate sensor data describing the exposure to one or more chemicals 904. In some embodiments, the sensor data can describe electrical signals (e.g., voltage or current) generated by the electrochemical sensors.

[0104] The generated sensor data 908 may then be processed by an embedding model 910 to produce an embedding output 912. The embedding model 910 may include one or more transformer models. In some embodiments, the embedding model 910 may include a graph neural network model 926, which may be trained to process both the graph representation 924 and the sensor data 908. Furthermore, the generated embedding 912 may be an embedding output of the embedding space and may include a set of identifier values ​​similar to RGB values ​​for color display.

[0105] In some embodiments, the system may be a bipedal system capable of processing either sensor data 908 or data describing a graph representation 924 to generate an embedded output 912. Furthermore, in some embodiments, a graph neural network model 926 and an embedding model 910 can be trained together. In some embodiments, the graph representation data 924 may be processed by the graph neural network model 926 before being processed by the embedding model 910. However, in some embodiments, the GNN model 926 may output an embedding that can be processed by a classification head 914 to determine a predictive characteristic label 916 without being processed by the embedding model 910.

[0106] Next, the generated embeddings 912 are processed by the classification head 914 to determine one or more matching predicted characteristic labels 916. The predicted characteristic labels 916 may include sensory characteristic labels such as smell, taste, or color. Then, the loss function 922 can be evaluated using the predicted characteristic labels 916 and the human-input characteristic labels 920. Next, the loss function 922 can be used to tune one or more parameters of at least one of the machine-learned models 910 and / or 926 by backpropagating the loss to learn / optimize the model parameters 918.

[0107] Process 900 can be iteratively completed over multiple training examples to train the machine learning models 910 and 926 to perform a classification task or generate an embedded output 912 that can be used to perform other tasks based on acquired sensor data 908.

[0108] Exemplary Method Figure 6 shows a flowchart illustrating an exemplary method performed by an exemplary embodiment of the present disclosure. While Figure 6 shows steps performed in a specific order for illustrative and illustrative purposes, the methods of the present disclosure are not limited to the illustrated order or arrangement. Various steps of Method 600 can be omitted, rearranged, combined, and / or adapted in various ways without departing from the scope of the present disclosure.

[0109] In 602, the computing system can generate sensor data. Sensor data can be generated using one or more sensors, which may include electrochemical sensors. In some embodiments, sensor data can describe electrical signals (e.g., voltage or current) generated by the sensors in response to exposure to one or more molecules.

[0110] In 604, the computing system can process sensor data using a machine learning model. The machine learning model may include one or more transformer models and / or one or more GNN embedding models. Furthermore, the machine learning model may be a machine learning model that has been trained to process sensor data and generate an embedded output in the embedding space.

[0111] In version 606, the computing system can generate embedded output. This embedded output may contain one or more values ​​similar to the RGB values ​​of a color display.

[0112] In 608, the computing system can perform tasks based on the embedded output. For example, the embedded output can be processed by a classification model to determine the sensed chemical or the characteristics of the sensed chemical. Classification of the embedded output may include the use of labeled embeddings in the embedding space, training examples, or other classification methods. In some embodiments, the embedded output may be processed by a classification head to determine the sensory characteristics of the sensed chemical (e.g., smell, taste, color). In other embodiments, the classification head may be trained to identify disease conditions based on the embedded output. The embedded output may be used to enable a sensor device to identify food spoilage, diseased crops, foul odors, etc., in real time.

[0113] Figure 7 shows a flowchart illustrating an exemplary method performed by an exemplary embodiment of the present disclosure. While Figure 7 shows steps performed in a specific order for illustrative and illustrative purposes, the methods of the present disclosure are not limited to the illustrated order or arrangement. Various steps of Method 700 can be omitted, rearranged, combined, and / or adapted in various ways without departing from the scope of the present disclosure.

[0114] In 702, the computing system can acquire sensor data. Sensor data can be acquired from one or more sensors and can describe exposure to one or more molecules.

[0115] In 704, the computing system can process sensor data using a machine learning model. The machine learning model may include one or more embedding models that are trained to process sensor data describing raw electrical signals and generate embedded outputs.

[0116] In 706, the computing system can generate embedded output.

[0117] In 708, the computing system can process the embedding output using a classification model to determine the classification. The classification model may include one or more classification heads trained to identify one or more matching labels in the embedding space. In some embodiments, the classification model may determine the label associated with the embedding output based on the similarity of a threshold determined at least partially at the value of the embedding output or at the location of the embedding output in the embedding space.

[0118] In 710, the computing system can provide classifications for display. Classifications may include the identification of chemical mixtures, the prediction of one or more properties, or other forms of classification (e.g., classification of disease conditions, classification of food spoilage, classification of ripeness, classification of malodorous odors, classification of diseased crops, etc.). Displays may include LED displays, LCD displays, ELD displays, plasma displays, QLED displays, or one or more lights attached to labels. In some embodiments, classifications may be displayed along with a visual representation of the embedded output in the embedded space. Furthermore, in some embodiments, similarity scores for different classifications may be displayed. If no classification meets a threshold, the system may display the closest class along with its similarity score.

[0119] Figure 8 shows a flowchart illustrating an exemplary method performed by an exemplary embodiment of the present disclosure. While Figure 8 shows steps performed in a specific order for illustrative and illustrative purposes, the methods of the present disclosure are not limited to the illustrated order or arrangement. Various steps of Method 800 can be omitted, rearranged, combined, and / or adapted in various ways without departing from the scope of the present disclosure.

[0120] In 802, the computing system can obtain examples of compound training. These compound training examples may include electrical signal training data and their respective training labels. The electrical signal training data and their respective training labels can describe a specific training compound.

[0121] In 804, the computing system can process training electrical signal data using a machine learning model to generate a compound embedding output. The compound embedding output may include embeddings of the embedding space.

[0122] In 806, the computing system can process the embedded output of compounds using a classification model to determine the compound label. The classification model can be trained to identify one or more relevant compound labels. In some embodiments, the classification model may include one or more classification heads trained for specific classifications.

[0123] In 808, the computing system can evaluate a loss function that assesses the difference between the compound label and its respective training label.

[0124] In 810, the computing system can tune one or more parameters of the machine-trained model based at least partially on the loss function.

[0125] Additional disclosures The technologies described herein refer to servers, databases, software applications, and other computer-based systems, as well as the actions performed and the information transmitted to and from such systems. The inherent flexibility of computer-based systems enables a wide variety of possible configurations, combinations, and divisions of tasks and functions among components. For example, the processes described herein can be implemented using a single device or component, or multiple devices or components working together. Databases and applications can be implemented in a single system or distributed across multiple systems. Distributed components can operate sequentially or in parallel.

[0126] While the subject matter of the present invention has been described in detail with respect to various specific exemplary embodiments thereof, each example is presented for illustrative purposes only and does not limit the disclosure. Those skilled in the art, having understood the foregoing, will readily be able to generate modifications, variations, and equivalents to such embodiments. Therefore, this disclosure does not preclude the inclusion of such modifications, variations, and / or additions to the subject matter that would be readily apparent to those skilled in the art. For example, features illustrated or described as part of one embodiment can be used in conjunction with another embodiment to obtain yet another embodiment. Thus, this disclosure is intended to cover such modifications, variations, and equivalents.

Claims

[Claim 1] The invention described herein.

Images