Explainable Machine Learning via Wavelet Analysis on Neural Network Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models, particularly black-box models like artificial neural networks, struggle to provide explainable outcomes, making it difficult to justify, track, or verify their results and improve the models based on these outcomes.
Innovation Solution
The use of wavelet analysis on the parameters of a machine learning model, specifically a convolutional neural network-based risk prediction model, to generate explanatory data for risk indicators, thereby providing insights into the features or characteristics contributing to the predicted outcomes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models with intricate structures (e.g., artificial neural networks) are used to provide high accuracy predictions, then prediction accuracy is improved, but model interpretability deteriorates
Solution Approach 1:
The patent segments the neural network parameters into distinct groups (e.g., weight matrices, bias vectors, activation functions) and applies wavelet transform to each segment separately. This segmentation allows the complex model to be broken down into interpretable components while preserving the overall predictive accuracy of the full model.
Solution Approach 2:
The wavelet transform acts as an intermediary between the neural network parameters and the interpretability requirement. It transforms the abstract parameter space into a domain where temporal-frequency characteristics are explicit, providing a bridge that maintains prediction accuracy while enabling explanation of model decisions.
2Loss of information
If wavelet transformation is applied to neural network parameters to generate explanatory data, then model interpretability is improved, but computational complexity increases
Solution Approach 1:
The patent changes the parameter representation from raw neural network weights to wavelet coefficients. By transforming parameters into the wavelet domain, the system maintains interpretability while the computational overhead is managed through efficient wavelet algorithms that can be applied to the parameter matrices without requiring complete model retraining.
Data Source
AI summary
A method can be used to predict risk and provide explainable outcomes using machine learning based on wavelet analysis. A risk prediction model can be applied to time-series data for an attribute associated with a target entity to generate a risk indicator for the target entity. The risk prediction model can include a feature learning model and a risk classification model configured to generate the risk indicator as output. Parameters of the feature learning model can be accessed and a plurality of basis functions of a wavelet transformation can be applied on the parameters of the feature learning model to generate a set of parameter wavelet coefficients. Explanatory data can be generated for the risk indicator based on the set of parameter wavelet coefficients. A responsive message can be transmitted to a remote computing device including the risk indicator and the explanatory data for use in controlling access of the target entity to an interactive computing environment.