Latent space harmonization for predictive modeling

Abstract

In embodiments of latent space harmonization (LSH) for predictive modeling, different training data sets are obtained from different measurement methods, where input data among the training data sets is quantifiable in a common space but a mapping between output data among the training data sets is unknown. A LSH module receives the training data sets and maps a common supervised target variable of the output data to a shared latent space where the output data can be jointly yielded. Mappings from the shared latent space back to the output training data of each training data set are determined and used to generate a trained predictive model. The trained predictive model is useable to predict output data from new input data with improved predictive power from the training data obtained using various, otherwise incongruent, measurement techniques.

1 - 20 . (canceled) 
     
     
         21 . A computing system comprising:
 a processing circuitry;   a memory including instructions that, when executed by the processing circuitry, cause the processing to perform operations comprising:   receiving a trained predictive model, the trained predictive model trained based on a first mapping function that maps a first supervised variable to a shared latent space and a second mapping function that maps a second supervised variable to a shared latent space, the first and second supervised variables representing respective incomparable measurement spaces;   receiving new data; and   operating the trained predictive model on the new data to generate output data, the output data representative of using a measurement technique of a measurement space of the measurement spaces on the new data.   
     
     
         22 . The computing system of  claim 21 , wherein the operations further comprise:
 transforming output data generated by the trained predictive model in the shared latent space to output data in a measurement scale of a measurement space of the measurement spaces.   
     
     
         23 . The computing system of  claim 21 , wherein the operations further comprise:
 receiving training data sets of input training data and output training data for each of the measurement spaces, the output data of each of the measurement spaces including a respective supervised target variable.   
     
     
         24 . The computing system of  claim 21 , wherein the input training data shares a common measurement scale. 
     
     
         25 . The computing system of  claim 21 , wherein output training data for a first measurement space of the measurement spaces is quantified in a measurement scale that is different from output training data for a second measurement space of the measurement spaces. 
     
     
         26 . The computing system of  claim 21 , wherein each of the measurement spaces corresponds to a different measurement technique. 
     
     
         27 . The computing system of  claim 25 , wherein the output training data in the shared latent space comprises a single data set that is mappable to supervised target variables included in each of the two or more measurement groups, and pairs of input data and output data in the shared latent space preserves an ordering of corresponding pairs of input training data and output training data in the training data sets. 
     
     
         28 . A computer-implemented method comprising:
 receiving a trained predictive model, the trained predictive model trained based on a first mapping function that maps a first supervised variable to a shared latent space and a second mapping function that maps a second supervised variable to a shared latent space, the first and second supervised variables representing respective incomparable measurement spaces;   receiving new data; and   operating the trained predictive model on the new data to generate output data, the output data representative of using a measurement technique of a measurement space of the measurement spaces on the new data.   
     
     
         29 . The method of  claim 28 , wherein the method further comprises:
 transforming output data generated by the trained predictive model in the shared latent space to output data in a measurement scale of a measurement space of the measurement spaces.   
     
     
         30 . The method of  claim 28 , wherein the method further comprises:
 receiving training data sets of input training data and output training data for each of the measurement spaces, the output data of each of the measurement spaces including a respective supervised target variable.   
     
     
         31 . The method of  claim 28 , wherein the input training data shares a common measurement scale. 
     
     
         32 . The method of  claim 28 , wherein output training data for a first measurement space of the measurement spaces is quantified in a measurement scale that is different from output training data for a second measurement space of the measurement spaces. 
     
     
         33 . The method of  claim 28 , wherein each of the measurement spaces corresponds to a different measurement technique. 
     
     
         34 . The method of  claim 32 , wherein the output training data in the shared latent space comprises a single data set that is mappable to supervised target variables included in each of the two or more measurement groups, and pairs of input data and output data in the shared latent space preserves an ordering of corresponding pairs of input training data and output training data in the training data sets. 
     
     
         35 . A non-transitory computer-readable storage device including instructions that, when executed by a machine, cause the machine to perform operations comprising:
 receiving a trained predictive model, the trained predictive model trained based on a first mapping function that maps a first supervised variable to a shared latent space and a second mapping function that maps a second supervised variable to a shared latent space, the first and second supervised variables representing respective incomparable measurement spaces;   receiving new data; and   operating the trained predictive model on the new data to generate output data, the output data representative of using a measurement technique of a measurement space of the measurement spaces on the new data.   
     
     
         36 . The non-transitory computer-readable storage device of  claim 35 , wherein the method further comprises:
 transforming output data generated by the trained predictive model in the shared latent space to output data in a measurement scale of a measurement space of the measurement spaces.   
     
     
         37 . The non-transitory computer-readable storage device of  claim 35 , wherein the method further comprises:
 receiving training data sets of input training data and output training data for each of the measurement spaces, the output data of each of the measurement spaces including a respective supervised target variable.   
     
     
         38 . The non-transitory computer-readable storage device of  claim 35 , wherein the input training data shares a common measurement scale. 
     
     
         39 . The non-transitory computer-readable storage device of  claim 35 , wherein output training data for a first measurement space of the measurement spaces is quantified in a measurement scale that is different from output training data for a second measurement space of the measurement spaces. 
     
     
         40 . The non-transitory computer-readable storage device of  claim 35 , wherein each of the measurement spaces corresponds to a different measurement technique.