Learning semantic description of data based on physical knowledge data

1 . A method for learning semantic description of data based on physical knowledge in a computing environment, comprising:

learning, by a processor, physical knowledge data and semantic labels associated with data from one or more data sources, wherein the processor is one of a plurality of components, wherein the plurality of components further include a physical knowledge component, a semantic label component, an equation discovery component, and a machine learning model component;

associating, by the processor, source attributes of the one or more data sources with one or more classes and concepts of a plurality of ontologies based on the physical knowledge data and the semantic labels to generate textual descriptors of the data;

ranking, by the processor, each of the one or more classes and concepts of the plurality of ontologies according to a degree of similarities between the source attributes of the one or more data sources and the data based on the physical knowledge data and the semantic labels; and

generating, using a plurality of layers, a semantic descriptor for each of the one or more data sources, wherein the plurality of layers include at least a platform layer, a machine learning layer, and a services layer, wherein the platform layer processes the data for the machine learning layer using one or more data cleansing frameworks, wherein pattern extractions occur between the machine learning layer and the services layer where the machine learning layer prioritizes one or more physical equations based on a match between the one or more physical equations and the data, and wherein the services layer enables a user to interface with a trained model.

2 . The method of claim 1 , further comprising:

identifying, by an equation discovery model, one or more physical equation models having a degree of relevancy related to the data; and

identifying, by the equation discovery model, the semantic labels from one or more physical equation models having a degree of relevancy to the data.

3 . The method of claim 1 , further comprising:

receiving multidimensional data that is time dependent;

applying a machine learning operation that combines embedding learning with self-supervised learning to identify one or more function labels that describe the dataset; and

identifying one or more physical equation models based on the one or more function labels by mapping variables in a function to concepts using a graph convolutional neural network to embed a mathematical expression into low-dimensional vector spaces.

4 . The method of claim 1 , further including initializing a machine learning model to:

translate time series data into one or more physical equation models representing and describing behavior of the physical knowledge data and the semantic labels, wherein the physical knowledge data includes one or more rules, policies, and laws pertaining to physics, and the one or more physical equation models and semantic labels are used by one or more automated feature engineering models; and

execute one or more prediction and interpolations operations based on the dataset.

5 . A system for learning semantic description of data based on physical knowledge in a computing environment, comprising:

one or more computers with executable instructions that when executed cause the system to:

learn physical knowledge data and semantic labels associated with data from one or more data sources, wherein the processor is one of a plurality of components, wherein the plurality of components further include a physical knowledge component, a semantic label component, an equation discovery component, and a machine learning model component;

associate source attributes of the one or more data sources with one or more classes and concepts of a plurality of ontologies based on the physical knowledge data and the semantic labels;

rank each of the one or more classes and concepts of the plurality of ontologies according to a degree of similarities between the source attributes of the one or more data sources and the data based on the physical knowledge data and the semantic labels; and

generate, using a plurality of layers a semantic descriptor for each of the one or more data sources, wherein the plurality of layers include at least a platform layer, a machine learning layer, and a services layer, wherein the platform layer processes the data for the machine learning layer using one or more data cleansing frameworks, wherein pattern extractions occur between the machine learning layer and the services layer where the machine learning layer prioritizes one or more physical equations based on a match between the one or more physical equations and the data, and wherein the services layer enables a user to interface with a trained model.

6 . The system of claim 5 , wherein the executable instructions that when executed cause the system to:

identify, by an equation discovery model, one or more physical equation models having a degree of relevancy related to the data; and

identify, by the equation discovery model, the semantic labels from one or more physical equation models having a degree of relevancy to the data.

7 . The system of claim 5 , wherein the executable instructions that when executed cause the system to:

receive multidimensional data that is time dependent;

apply a machine learning operation that combines embedding learning with self-supervised learning to identify one or more function labels that describe the dataset; and

identify one or more physical equation models based on the one or more function labels by mapping variables in a function to concepts using a graph convolutional neural network to embed a mathematical expression into low-dimensional vector spaces.

8 . The system of claim 5 , wherein the executable instructions that when executed cause the system to initialize a machine learning model to:

translate time series data into one or more physical equation models representing and describing behavior of the physical knowledge data and the semantic labels, wherein the physical knowledge data includes one or more rules, policies, and laws pertaining to physics, and the one or more physical equation models and semantic labels are used by one or more automated feature engineering models; and

execute one or more prediction and interpolations operations based on the dataset.

9 . A computer program product for learning semantic description of data based on physical knowledge in a computing environment, the computer program product comprising:

one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instruction comprising:

program instructions to learn physical knowledge data and semantic labels associated with data from one or more data sources, wherein the processor is one of a plurality of components, wherein the plurality of components further include a physical knowledge component, a semantic label component, an equation discovery component, and a machine learning model component;

program instructions to associate source attributes of the one or more data sources with one or more classes and concepts of a plurality of ontologies based on the physical knowledge data and the semantic labels;

program instructions to rank each of the one or more classes and concepts of the plurality of ontologies according to a degree of similarities between the source attributes of the one or more data sources and the data based on the physical knowledge data and the semantic labels; and

program instructions to generate, using a plurality of layers, a semantic descriptor for each of the one or more data sources, wherein the plurality of layers include at least a platform layer, a machine learning layer, and a services layer, wherein the platform layer processes the data for the machine learning layer using one or more data cleansing frameworks, wherein pattern extractions occur between the machine learning layer and the services layer where the machine learning layer prioritizes one or more physical equations based on a match between the one or more physical equations and the data, and wherein the services layer enables a user to interface with a trained model.

10 . The computer program product of claim 9 , further including program instructions to:

identify, by an equation discovery model, one or more physical equation models having a degree of relevancy related to the data; and

identify, by the equation discovery model, the semantic labels from the one or more physical equation models having a degree of relevancy to the data.

11 . The computer program product of claim 9 , further including program instructions to:

receive multidimensional data that is time dependent;

apply a machine learning operation that combines embedding learning with self-supervised learning to identify one or more function labels that describe the dataset; and

identify one or more physical equation models based on the one or more function labels by mapping variables in a function to concepts using a graph convolutional neural network to embed a mathematical expression into low-dimensional vector spaces.

12 . The computer program product of claim 9 , further including program instructions to initialize a machine learning model to:

translate time series data into one or more physical equation models representing and describing behavior of the physical knowledge data and the semantic labels, wherein the physical knowledge data includes one or more rules, policies, and laws pertaining to physics, and the one or more physical equation models and semantic labels are used by one or more automated feature engineering models; and

execute one or more prediction and interpolations operations based on the dataset.

13 . The method of claim 2 , further comprising:

generating, by the semantic label component, formula embeddings for each of the one or more physical equation models;

identifying, by the semantic label component, similar embeddings based on a similarity measure;

ranking, by the semantic label component, the similar embeddings based on one or more confidence metrics;

retrieving, by the semantic label component, a plurality of ranked formulas; and

retrieving, by the semantic label component, source data associated with each of the plurality of ranked formulas, wherein the source data identifies textual descriptors for each of the plurality of ranked formulas.

14 . The method of claim 1 , wherein a machine learning model of the machine learning component may be trained based on the physical knowledge data and the semantic labels associated with the data from the one or more data sources and performs the associating of the source attributed of the one or more data sources with the one or more classes and concepts of the plurality of ontologies.

15 . The method of claim 1 , further comprising:

processing, by the physical knowledge component, each sensor signal received from one or more Internet of Things (IoT) devices;

identifying, by the equation discovery component, one or more equations associated with the sensor signal processed by the physical knowledge component, wherein the terms of the one or more equations are directly computed from sensor signal data using finite difference methods and local coefficients; and

applying the one or more equations in a loss function of a neural net training algorithm.

16 . The method of claim 1 , wherein the plurality of layers use the physical knowledge data learned to guide feature engineering in the computing environment.

17 . The method of claim 1 , wherein the services layer provides specified monitoring of one or more machine learning models, including performance improvement of the one or more machine learning models and lifecycle management.

18 . The system of claim 5 , wherein the plurality of layers use the physical knowledge data learned to guide feature engineering in the computing environment, and wherein the services layer provides specified monitoring of one or more machine learning models, including performance improvement of the one or more machine learning models and lifecycle management.

19 . The computer program product of claim 9 , wherein the plurality of layers use the physical knowledge data learned to guide feature engineering in the computing environment, and wherein the services layer provides specified monitoring of one or more machine learning models, including performance improvement of the one or more machine learning models and lifecycle management.