Systems and methods for learning across multiple chemical sensing units using a mutual latent representation

A system comprising:

• at least one computer processor;
• a first chemical sensing unit;
• a plurality of chemical sensing units not including the first chemical sensing unit;
• and
• at least one non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by the at least one computer processor, cause the at least one computer processor to perform a method of training a set of models used by the first chemical sensing unit to generate an inference about one or more chemical characteristics of an environment, wherein the training uses information associated with the plurality of chemical sensing units, the method comprising:
• accessing the information associated with the plurality of chemical sensing units, wherein the information relates a plurality of signals output from the plurality of chemical sensing units to a mutual latent representation, wherein the mutual latent representation provides a common representation space for the plurality of signals output from the plurality of chemical sensing units;
• accessing a training dataset including values representing signals output from the first chemical sensing unit;
• and
• training the set of models associated with the first chemical sensing unit using the training dataset and the information associated with the plurality of chemical sensing units, wherein training the set of models comprises:
• training a first model of the set of models to relate the values representing the signals output from the first chemical sensing unit to a feature representation;
• training a second model of the set of models to relate the feature representation to a latent representation;
• and
• training a third model to relate the latent representation to the mutual latent representation,
• wherein the first chemical sensing unit is configured to use the trained first, second, and third models to generate an inference about one or more chemical characteristics of the environment based on the first chemical sensing unit sensing at least one analyte.

A chemical sensing system comprising:

• a first chemical sensing unit configured to:
• sense at least one analyte in a sample from an environment;
• and
• output signals generated as a result of sensing the at least one analyte;
• at least one computer processor;
• and
• at least one non-transitory computer-readable storage medium storing:
• information representing a set of models trained using information associated with a plurality of chemical sensing units not including the first chemical sensing unit;
• and
• processor-executable instructions that, when executed by the at least one computer processor, cause the at least one computer processor to:
• generate, using the information representing the set of models, an inference about one or more chemical characteristics of the environment based on the plurality of sensors sensing the at least one analyte in the sample by:
• providing the signals as input to a first model of the set of models, wherein the first model is trained to relate the signals to a feature representation to generate feature representation values;
• providing the feature representation values as input to a second model of the set of models, wherein the second model is trained to relate the feature representation to a latent representation to generate latent representation values;
• providing the latent representation values as input to a third model of the set of models, wherein the third model is trained to relate the latent representation to a mutual latent representation to generate mutual latent representation values, wherein the mutual latent representation provides a common representation space for signals output from multiple chemical sensing units;
• and
• providing the mutual latent representation values as input to a fourth model of the set of models, wherein the fourth model is trained to relate the mutual latent representation to inferences to generate the inference about the one or more chemical characteristics of the environment based on the plurality of sensors sensing the at least one analyte in the sample.

A method of generating an inference about one or more chemical characteristics of an environment by a chemical sensing system comprising a first chemical sensing unit and memory storing information representing a set of models trained using information associated with a plurality of chemical sensing units not including the first chemical sensing unit, the method comprising:

• sensing, by the first chemical sensing unit, at least one analyte in a sample from the environment;
• outputting, by the first chemical sensing unit, output signals generated as a result of sensing the at least one analyte;
• generating, using the information representing the set of models, the inference about one or more chemical characteristics of an environment based on the plurality of sensors sensing the at least one analyte in the sample by:
• providing the signals as input to a first model of the set of models, the first model trained to relate the signals to a feature representation to generate feature representation values;
• providing the feature representation values as input to a second model of the set of models, the second model trained to relate the feature representation to a latent representation to generate latent representation values;
• providing the latent representation values as input to a third model of the set of models, the third model trained to relate the latent representation to a mutual latent representation to generate mutual latent representation values, wherein the mutual latent representation provides a common representation space for signals output from multiple chemical sensing units;
• and
• providing the mutual latent representation values as input to a fourth model of the set of models, the fourth model trained to relate the mutual latent representation to inferences to generate the inference about the one or more chemical characteristics of the environment based on the plurality of sensors sensing the at least one analyte in the sample.