Methods, systems, and apparatuses for user-understandable explainable learning models

A system for constructing an explainable user output, the system comprising:

• a processor configured to receive input data to process a set of features at a neural network composed of a plurality of layers of an original classifier wherein the original classifier has been frozen with a set of weights related to features of the input data;
• the processor configured to determine a semantic function to categorize a data sample with a semantic category, and to determine a level of semantic accuracy for each layer of the plurality of layers of the original classifier within the neural network wherein the original classifier is a trained model;
• the processor configured to compute a representative vector with average activations of each layer's nodes of the plurality of layers for training a set of samples of a semantic category to evaluate by computation of distances of each sample in a test set composed of a set of samples from available layers of the plurality of layers for each semantic category of a set of semantic categories;
• the processor configured to compute a number of test samples for each layer of the plurality of layers and for each semantic category, that are closest to each other in each layer in the plurality of layers, to designate a layer of the plurality of layers with a highest score representative of a best semantics in each test samples wherein a designated layer is an optimal semantic content layer;
• the processor configured to extend the optimal semantic content layer by a category branch to the neural network to extract a set of semantic data samples from the semantic content of the original classifier to the neural network, and to extend with an extension the explainable classifier of the neural network to define a plurality of semantic categories;
• the processor configured to train a set of new connections of the explainable classifier of the neural network to compute a set of output explanations that are composed of explanations with an accuracy measure associated each output explanation based on at least one semantic category of the plurality of semantic categories;
• and
• the processor configured to compare the accuracy measure for each output explanation based on an extracted semantic data sample by the trained explainable classifier for each semantic category of the plurality of semantic categories and to generate the output explanation in a user understandable format.

An apparatus to execute an explainable classifier comprising:

• at least one processor deployed in a vehicle, the at least one processor programmed to:
• receive input data of a set of a plurality of features for processing at a neural network (NN) comprising one of a convoluted neural network (CNN) which is composed of a plurality of layers of an original classifier wherein the original classifier has been frozen with a set of weights related to features of the input data;
• determine a semantic function of a plurality of semantic functions to categorize a data sample with a semantic category;
• determine a level of semantic accuracy for each layer of the plurality of layers of the original classifier within the neural network wherein the original classifier is a trained model;
• compute a representative vector with average activations of each layer's nodes of the plurality of layers for training samples of a semantic category for evaluation by computing distances of samples in a test set from available layers for each semantic category;
• compute a number of test samples of a set of test samples for each layer and for each semantic category, that are closest to each other in each layer, to designate a layer of the plurality of layers with a highest score representative of a best semantics in each test samples;
• redefine the semantic function to change the semantic accuracy to generate more semantic content in determining the semantic accuracy of each layer of the original classifier by the trained model wherein the redefinition of the semantic function is an optional step to provide more abstract semantic categories;
• extend a designated layer by a category branch of the neural network to extract semantic data samples derived from the semantic content to the extension of the neural network, and to extend an explainable classifier of the neural network to define a plurality of semantic categories;
• train a set of connections of the explainable classifier of the neural network to compute a set of output explanations with an accuracy measure associated each output explanation based on at least one semantic category of the plurality of semantic categories wherein the output explanations are well defined syntactic sentences;
• and
• compare the accuracy measure for each output explanation based on an extracted semantic data sample by the trained explainable classifier for each semantic category to generate the output explanation in a user understandable format.