Neural Network Feature Selection via Partial Reconstruction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current machine learning models for fault diagnosis in devices face challenges in selecting relevant features effectively, leading to suboptimal prediction accuracy and robustness against changes in feature distributions.

Innovation Solution

A neural network architecture that includes a feature selection layer, a feature extraction layer, and a partial reconstruction layer, where weight parameters are adjusted based on prediction accuracy and reconstruction error, enabling the selection of features useful for prediction and improving generalization performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional feature selection methods are used in machine learning models, then the model can be trained with available features, but the prediction accuracy and robustness against changes in feature distributions deteriorate

Engineering Contradiction:

Improveprediction accuracyVSAvoidrobustness against changes in feature distributions

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The neural network is segmented into distinct functional layers: feature selection layer, feature extraction layer, and partial reconstruction layer. Each layer performs a specific function, allowing the system to optimize feature selection separately from feature extraction and prediction, thereby improving both accuracy and robustness

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The partial reconstruction layer reconstructs the selected features from extracted feature quantities and feeds this information back to the feature selection layer. This feedback mechanism allows the system to learn which features are most useful for prediction while maintaining robustness by comparing reconstructed versus original selected features

Inventive Principle:

Principle #23Feedback

2Loss of information

If all input features are used in the neural network, then comprehensive information is available for prediction, but the model complexity and computational cost increase

Engineering Contradiction:

Improveinformation availability for predictionVSAvoidmodel complexity

Core Design Contradiction:

Loss of informationVSDevice complexity

Solution Approach 1:

The feature selection layer extracts and selects only the most relevant features from the complete input feature set. By taking out only the necessary features rather than using all features, the system maintains information availability for prediction while reducing model complexity and computational requirements

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The partial reconstruction layer reconstructs only the selected features rather than all input features. This partial action approach maintains the essential information needed for prediction while avoiding the computational burden of processing and reconstructing the complete feature set

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20240037388A1Method of learning neural network, feature selection apparatus, feature selection method, and recording medium

Publication Date: 2024.02.01 NEC CORP

AI summary

A method of learning a neural network, wherein the neural network includes: a feature selection layer for selecting a part of input data; a feature extraction layer for extracting a feature quantity on the basis of the selected input data; a prediction layer for performing a prediction on the basis of the feature quantity; and a partial reconstruction layer for reconstructing the selected input data on the basis of the feature quantity, and the method includes adjusting a weight parameter of the neural network on the basis of a prediction accuracy by the prediction layer and a reconstruction error in the partial reconstruction layer.