Artificial Neural Network Weight Matrix Optimization via Marginal Contribution Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Artificial neural networks (ANNs) face challenges in explaining decision-making processes, leading to difficulties in improving decision-making accuracy and minimizing false positives, as they typically provide answers without reasoning.
Innovation Solution
A method and system that generate weight matrices for each ANN layer, determine marginal contribution values of neural nodes, execute elimination decisions based on these values, and update weights to reduce false positives by employing cooperative game theory algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional ANN methods are used to provide decision answers, then computation speed is maintained, but decision-making explainability and accuracy improvement capability deteriorate
Solution Approach 1:
The patent segments the ANN decision-making process into multiple interpretable components: input layer, hidden layers with weighted connections, and output layer. Each neural node's contribution is separately evaluated through marginal contribution analysis, allowing the system to explain decisions by tracing which specific nodes and connections influenced the outcome, thus resolving the contradiction between maintainability and complexity.
Solution Approach 2:
The patent introduces an intermediary explanation layer that mediates between the complex internal ANN computations and the final decision output. This layer generates human-readable explanations by analyzing marginal contributions of neural nodes and distributing surplus values, making the black-box ANN decisions transparent without requiring changes to the underlying computation engine, thereby resolving the contradiction between maintainability and complexity.
2Reliability
If more neural nodes are retained in each layer, then decision-making accuracy is improved, but computation time and resource consumption increase
Solution Approach 1:
The patent dynamically changes the effective parameters (weights) of neural nodes based on their marginal contribution values. Nodes with low or negative marginal contributions have their weights reduced or eliminated, while nodes with high positive contributions maintain or increase their weights. This parameter adjustment optimizes the network's decision-making accuracy by retaining only the most important nodes, thereby resolving the contradiction between reliability and productivity.
Solution Approach 2:
The patent applies local quality optimization by treating each neural node individually based on its specific contribution to the decision. Instead of uniformly processing all nodes, the system identifies and emphasizes locally important nodes (those with high marginal contributions) while reducing or eliminating locally unimportant ones. This localized optimization improves overall decision accuracy without requiring the entire network to process all nodes at full capacity, thus resolving the contradiction between reliability and productivity.
3Measurement precision
If weight updates are performed frequently during training, then learning accuracy is improved, but training time and computational resources increase
Solution Approach 1:
The patent implements periodic weight updates based on milestone achievements rather than continuous updates. The system performs comprehensive marginal contribution analysis and weight redistribution at specific intervals (e.g., after processing certain batches of data or achieving specific performance thresholds). This periodic action maintains learning accuracy by ensuring weights are updated when meaningful patterns are learned, while reducing unnecessary computational overhead from continuous updates, thus resolving the contradiction between measurement precision and loss of time.
Solution Approach 2:
The patent performs preliminary marginal contribution analysis and weight optimization before final decision-making or before major training milestones. By pre-calculating which neural nodes are most important and adjusting their weights in advance, the system achieves high learning accuracy without requiring frequent interruptive updates during the main training process, thereby resolving the contradiction between measurement precision and loss of time.
Data Source
AI summary
This disclosure relates to method and system for improving performance of an artificial neural network (ANN). The method may include generating a weight matrix comprising weights of neural nodes in a given layer for each layer of the ANN, determining a marginal contribution value of a given neural node for each neural node in the given layer with respect to other neural nodes in the given layer, executing an elimination decision for each neural node in each layer based on the corresponding marginal contribution value, determining a distributed surplus value of a given remaining neural node in a given layer based on the marginal contribution values of a coalition of remaining neural nodes in the given layer for each remaining neural node in each layer, and updating the weight matrix based on the distributed surplus value of each remaining neural node in each layer.