A deep learning model is proposed for finding human-understandable connections between input features. The approach uses a parameterized, differentiable activation function, based on the theoretical background of nilpotent fuzzy logic and multi-criteria decision-making. The learnable parameter has a semantic meaning indicating the level of compensation between input features. The neural network determines the parameters using gradient descent to find human-understandable relationships between input features. The utility and effectiveness of the model is demonstrate by successfully applying it to classification problems from the UCI Machine Learning Repository.
Code and details for the manuscript on human-understandable neural models. The repository should have everything needed to reproduce the results of the paper and get started exploring interptretability in other systems. All the Python codes should basically be self-contained, provided the dependencies are met.
Dependecies:
- Tensorflow: optimizers for training machine learning models
- Pandas: quantitative data analysis and manipulation tool
- Standard libraries: numpy, enum, tuple, math, time, json, datetime, sequence, sys, union, mapping, list, dict, random, os, csv, copy, any, optional, cast, textIO