Automatically reducing machine learning model inputs

Abstract

Various embodiments are generally directed to techniques to reduce inputs of a machine learning model (MLM) and increase path efficiency as a result. A method for reducing an MLM includes: receiving a machine learning (ML) dataset, partitioning the ML dataset into a first dataset, a second dataset, a third dataset, and a fourth dataset, training, validating, and testing the MLM using one or more of the first dataset, the second dataset, and the third dataset, after testing the MLM, automatically ranking an importance associated with each input of the MLM using the fourth dataset, and reducing a plurality of inputs of the MLM based on the automatic ranking.

What is claimed is: 
     
         1 . A method, comprising:
 generating, by a computer processor, a reduced input machine learning model (MLM) based on training an original MLM, wherein generating the reduced input MLM comprises:
 training the original MLM using a first dataset, to generate a plurality of weights for the original MLM; 
 pruning the reduced input MLM such that a first input of a plurality of inputs of the original MLM is not included as an input of the reduced input MLM; 
 determining that a size of the reduced input MLM exceeds a memory threshold of a target device; and 
 based on the determination that the size of the reduced input MLM is greater than the memory threshold of the target device, prune the reduced input MLM to remove a second input of the plurality of inputs of the original MLM included in the reduced input MLM to generate a second reduced MLM; and 
   processing, by the second reduced MLM executing on the processor, an applied dataset.   
     
     
         2 . The method of  claim 1 , wherein pruning the reduced input MLM comprises:
 automatically ranking each of the plurality of inputs of the original MLM.   
     
     
         3 . The method of  claim 2 , wherein the ranking is based on a multimodal gaussian distribution that includes three or more peaks, wherein the multimodal gaussian distribution is generated based on a variant analysis during a backpropagation operation of each input of the original MLM, wherein the reduction of the plurality of inputs is based on one or more results of the variant analysis. 
     
     
         4 . The method of  claim 3 , wherein the variant analysis of the backpropagation operation sums each weight along a path of each input of the original MLM, wherein the first input is removed based on the summed weights of each path. 
     
     
         5 . The method of  claim 4 , wherein the variant analysis further comprises normalizing each summed weight. 
     
     
         6 . The method of  claim 1 , further comprising:
 comparing the size of the reduced input MLM to the memory threshold to determine the reduced input MLM exceeds the memory threshold of the target device.   
     
     
         7 . The method of  claim 1 , wherein the memory threshold comprises a memory capacity of the target device. 
     
     
         8 . A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a processor, cause the processor to:
 generate a reduced input machine learning model (MLM) based on training an original MLM, wherein generating the reduced input MLM comprises:
 training the original MLM using a first dataset, to generate a plurality of weights for the original MLM; 
 pruning the reduced input MLM such that a first input of a plurality of inputs of the original MLM is not included as an input of the reduced input MLM; 
 determining that a size of the reduced input MLM exceeds a memory threshold of a target device; and 
 based on the determination that the size of the reduced input MLM is greater than the memory threshold of the target device, pruning the reduced input MLM to remove a second input of the plurality of inputs of the original MLM included in the reduced input MLM to generate a second reduced MLM; and 
   process, by the second reduced MLM executing on the processor, an applied dataset.   
     
     
         9 . The computer-readable storage medium of  claim 8 , wherein pruning the reduced input MLM comprises:
 automatically rank each of the plurality of inputs of the original MLM.   
     
     
         10 . The computer-readable storage medium of  claim 9 , wherein the ranking is based on a multimodal gaussian distribution that includes three or more peaks, wherein the multimodal gaussian distribution is generated based on a variant analysis during a backpropagation operation of each input of the original MLM, wherein the reduction of the plurality of inputs is based on one or more results of the variant analysis. 
     
     
         11 . The computer-readable storage medium of  claim 10 , wherein the variant analysis of the backpropagation operation sums each weight along a path of each input of the original MLM, wherein the first input is removed based on the summed weights of each path. 
     
     
         12 . The computer-readable storage medium of  claim 11 , wherein the variant analysis further comprises normalize each summed weight. 
     
     
         13 . The computer-readable storage medium of  claim 8 , wherein the instructions further cause the processor to:
 compare the size of the reduced input MLM to the memory threshold to determine the reduced input MLM exceeds the memory threshold of the target device.   
     
     
         14 . The computer-readable storage medium of  claim 8 , wherein the memory threshold comprises a memory capacity of the target device. 
     
     
         15 . A computing apparatus comprising:
 a processor; and   a memory storing instructions that, when executed by the processor, cause the processor to:
 generate a reduced input machine learning model (MLM) based on training an original MLM, wherein generating the reduced input MLM comprises:
 training the original MLM using a first dataset, to generate a plurality of weights for the original MLM; 
 pruning the reduced input MLM such that a first input of a plurality of inputs of the original MLM is not included as an input of the reduced input MLM; 
 determining that a size of the reduced input MLM exceeds a memory threshold of a target device; and 
 based on the determination that the size of the reduced input MLM is greater than the memory threshold of the target device, pruning the reduced input MLM to remove a second input of the plurality of inputs of the original MLM included in the reduced input MLM to generate a second reduced MLM; and 
 
 process, by the second reduced MLM executing on the processor, an applied dataset. 
   
     
     
         16 . The computing apparatus of  claim 15 , wherein pruning the reduced input MLM comprises:
 automatically rank each of the plurality of inputs of the original MLM.   
     
     
         17 . The computing apparatus of  claim 16 , wherein the ranking is based on a multimodal gaussian distribution that includes three or more peaks, wherein the multimodal gaussian distribution is generated based on a variant analysis during a backpropagation operation of each input of the original MLM, wherein the reduction of the plurality of inputs is based on one or more results of the variant analysis. 
     
     
         18 . The computing apparatus of  claim 17 , wherein the variant analysis of the backpropagation operation sums each weight along a path of each input of the original MLM, wherein the first input is removed based on the summed weights of each path. 
     
     
         19 . The computing apparatus of  claim 18 , wherein the variant analysis further comprises normalize each summed weight. 
     
     
         20 . The computing apparatus of  claim 15 , wherein the memory threshold comprises a memory capacity of the target device, wherein the instructions further cause the processor to:
 compare the size of the reduced input MLM to the memory threshold to determine the reduced input MLM exceeds the memory threshold of the target device.