Hypergraph Feature Selection for Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current feature selection methods for machine learning models fail to consider learning objectives and relations based on feature values, leading to inefficiencies in training and increased complexity, particularly when dealing with multiple learning objectives.
Innovation Solution
A system that generates a hypergraph based on features and relations, using transitive closure to select a subset of features for each learning objective, considering multiple relations and feature values, thereby improving feature selection accuracy and reducing overfitting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional feature selection methods are used, then the process is simple, but they fail to consider learning objectives and relations based on feature values, leading to increased model complexity and overfitting
Solution Approach 1:
The patent segments the feature selection process into distinct components: (1) constructing a hypergraph structure that separates features, relations, and learning objectives into organized nodes and hyperedges, and (2) applying transitive closure algorithm to systematically traverse and select features based on learned relations. This segmentation allows the system to consider multiple learning objectives and feature relations without overwhelming complexity.
Solution Approach 2:
The patent introduces a hypergraph as an intermediary data structure that mediates between the input features/relations and the feature selection process. The hypergraph captures complex relations between features and learning objectives, allowing the system to systematically traverse these relations through transitive closure to select features that satisfy multiple learning objectives simultaneously.
2Reliability
If multiple learning objectives are considered in feature selection, then model performance improves, but the selection process becomes more complex and time-consuming
Solution Approach 1:
The patent performs preliminary action by pre-construction the hypergraph structure that encodes all relations between features and learning objectives before the actual feature selection occurs. This pre-processing step organizes the complex relationships in advance, allowing the transitive closure algorithm to efficiently traverse pre-established paths rather than computing relations on-the-fly during feature selection.
Solution Approach 2:
The patent creates a hypergraph copy or representation of the feature-relation-objective space that can be traversed independently. Instead of directly analyzing the complex original data relationships during feature selection, the system works with the hypergraph structure that copies and simplifies these relationships, making the selection process more efficient while maintaining accuracy.
3Measurement precision
If feature selection considers relations based on feature values, then selection accuracy improves, but computational complexity increases
Solution Approach 1:
The patent changes the parameter representation by transforming feature relations into a hypergraph structure where relations become first-class citizens represented as hyperedges connecting feature nodes. This parameter transformation allows the system to systematically traverse relations using graph algorithms rather than computing complex value-based relationships directly, reducing computational complexity while maintaining the ability to consider feature value relations.
Data Source
AI summary
An example system includes a processor to receive a set of features, a set of relations between the features, and a set of target features. Each of the target features is associated with a number of the relations. The processor can generate a hypergraph based on the features and the relations. The processor also can select a subset of features based on a transitive closure of the hypergraph for each of the target features. The processor can transmit the selected subset of features.