Automatically reducing machine learning model inputs

A method, comprising:

• 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, by a computer processor, a machine learning model (MLM) based on the first dataset, wherein the training includes generating a plurality of weights for the MLM;
• after training the MLM, validating at least one hyperparameter of the MLM based on the second dataset;
• after validating the MLM, testing i) an accuracy ability, ii) a precision ability, and iii) a recall ability using the third dataset;
• after testing the MLM, automatically ranking an importance associated with each of a plurality of inputs of the MLM using the fourth dataset, 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;
• reducing the plurality of inputs of the MLM based on the automatic ranking by removing a first input of the plurality of inputs as input to the MLM, wherein each of the first dataset, the second dataset, the third dataset, and the fourth dataset are distinct from one another;
• comparing a size of the reduced input MLM and a memory capacity of a target device to determine that the size of the reduced MLM is greater than the memory capacity of the target device;
• and
• based on the determination that the size of the reduced input MLM is greater than the memory capacity 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.

A non-transitory computer-readable storage medium storing computer-readable program code executable by a processor to:

• train, by a computer processor, a machine learning model (MLM) using a first dataset, wherein the training includes generating a plurality of weights for the MLM;
• validating at least one hyperparameter of the MLM using a second dataset;
• testing i) an accuracy ability, ii) a precision ability, and iii) a recall ability using a third dataset;
• automatically rank an importance associated with each of a plurality of inputs of the MLM using a fourth dataset, 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;
• reduce the plurality of inputs of the MLM based on the automatic ranking by removing a first input of the plurality of inputs as input to the MLM;
• after reducing the plurality of inputs of the MLM, retest the reduced input MLM using a fifth dataset, wherein the retesting includes testing i) an accuracy ability, ii) a precision ability, and iii) a recall ability of the reduced input MLM;
• compare a size of the reduced input MLM and a memory capacity of a target device to determine that the size of the reduced MLM is greater than the memory capacity of the target device;
• and
• based on the determination that the size of the reduced input MLM is greater than the memory capacity 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.