Intermediary Model for Interpretable Deep Neural Network Predictions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep neural networks (DNNs) struggle to provide interpretable results due to their complex architecture, making it difficult for marketers and users to understand the underlying factors influencing predictions, which can lead to a lack of trust and accuracy in AI-driven decision-making processes.
Innovation Solution
By training multiple machine learning models and a DNN on the same dataset, evaluating their performance using standard accuracy metrics and distribution analysis, and selecting the best-performing model to provide factors that explain the DNN's results, ensuring accuracy and confidence in AI predictions without requiring extensive computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional machine learning models are used to interpret results, then interpretability is improved, but prediction accuracy and performance deteriorate
Solution Approach 1:
The patent introduces an intermediary model (traditional machine learning model) that mediates between the black-box DNN and the user. The intermediary model is trained to replicate DNN predictions while maintaining interpretability, allowing users to understand predictions without sacrificing DNN accuracy. This resolves the contradiction by providing a transparent intermediate layer that bridges the gap between complex DNN performance and human-understandable explanations.
Solution Approach 2:
The patent creates a copy of the DNN's predictive behavior using a traditional machine learning model. By training the intermediary model to reproduce DNN outputs on the same dataset, the system captures the essential predictive patterns in an interpretable form. This copying approach allows the system to maintain DNN-level accuracy while providing transparent, explainable predictions through the simpler model structure.
2Productivity
If DNN is used to handle large amounts of data, then prediction performance is improved, but interpretability and understanding of underlying factors deteriorate
Solution Approach 1:
The intermediary traditional machine learning model serves as a mediator that translates the complex patterns learned by the DNN from large datasets into interpretable factors. The intermediary model is trained on the same large dataset but produces predictions that can be traced back to specific input features, thereby recovering the understandability lost in the DNN's complex processing of large-scale data.
3Loss of information
If methods alter input values to determine impact on output, then factor analysis is improved, but computational resources required increase extensively
Solution Approach 1:
The patent performs preliminary action by training the intermediary model once to capture the relationship between inputs and outputs. Instead of repeatedly altering input values during analysis, the intermediary model's trained parameters directly reveal the impact of different factors. This preliminary training phase eliminates the need for extensive computational experiments during the analysis phase, dramatically reducing ongoing computational resource requirements while maintaining factor analysis capability.
Data Source
AI summary
Embodiments of the present invention provide systems, methods, and computer storage media for providing factors that explain the generated results of a deep neural network (DNN). In embodiments, multiple machine learning models and a DNN are trained on a training dataset. A preliminary set of trained machine learning models with similar results to the trained DNN are selected for further evaluation. The preliminary set of machine learning models may be evaluated using a distribution analysis to select a reduced set of machine learning models. Results produced by the reduced set of machine learning models are compared, point-by-point, to the results produced by the DNN. The best performing machine learning model with generated results that performs closest to the DNN generated results may be selected. One or more factors used by the selected machine learning model are determined. Those one or more factors from the selected best performing machine learning model may be provided to explain the results of the DNN and increase confidence in the understanding and accuracy of the results generated by the DNN.