Biometric Recognition Model Parameter Optimization via LHS and NSGA
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-objective optimization methods for biometric recognition models struggle to optimize multiple objectives, such as recognition accuracy and latency, in a balanced manner, leading to inefficient acquisition of Pareto optimal solutions.
Innovation Solution
A method and device that utilize Latin Hypercube Sampling (LHS) and a Non-dominated Ranking Genetic Algorithm (NSGA) to select test data sets and update the decision space, enabling the determination of Pareto solution sets and improving optimization efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional multi-objective optimization method assigns weights to objective functions, then multiple objectives can be optimized, but optimization efficiency is low due to weight assignment constraints
Solution Approach 1:
The patent extracts the weight assignment constraint from the optimization process by using NSGA-III which inherently handles multiple objectives without requiring explicit weight assignments. This removes the problematic element (weight assignment) while preserving the multi-objective optimization capability.
Solution Approach 2:
The patent replaces the mechanical weight assignment system with an evolutionary algorithm-based selection mechanism. Instead of manually assigning weights to objective functions, the system uses NSGA-III's built-in reference point mechanism and evolutionary selection to automatically balance multiple objectives, substituting a more efficient computational approach.
2Productivity
If intelligent multi-objective optimization method uses genetic algorithms to take samples, then optimization efficiency is improved, but respective objectives cannot be effectively optimized in a balanced manner in partial decision space
Solution Approach 1:
The patent introduces reference points in the objective space as an additional dimensional structure to guide the evolutionary search. This reference point mechanism adds a new dimension to the optimization process, enabling balanced exploration of the decision space while maintaining high optimization efficiency through the genetic algorithm framework.
Solution Approach 2:
The patent changes the parameter representation by using reference points and diversity maintenance mechanisms in NSGA-III. Instead of relying solely on traditional genetic algorithm parameters, the system incorporates reference point parameters that guide the search toward balanced objective optimization, thereby improving both efficiency and balance.
3Loss of time
If evaluation is performed only in partial decision space, then computational cost is reduced, but optimal solution set cannot be found quickly due to unbalanced objective optimization
Solution Approach 1:
The patent performs preliminary action by pre-defining reference points in the objective space before the optimization process begins. These reference points serve as guideposts that direct the evolutionary search toward regions of the decision space that are likely to contain balanced optimal solutions, reducing the time needed to discover the optimal solution set.
Solution Approach 2:
The patent implements feedback through the reference point mechanism and diversity maintenance in NSGA-III. The algorithm continuously monitors the distribution of solutions relative to reference points and adjusts the search direction accordingly, providing feedback that accelerates the discovery of optimal solutions while maintaining balance across multiple objectives.
Data Source
AI summary
A method and device for optimizing parameters of a biometric recognition mode are provided, the method includes: repeating steps of selecting a first test data set SI from current decision space based on Latin Hypercube Sampling (LHS), obtaining a first set O1 of multi-objective values corresponding to the set S1 determined based on original evaluation and determining a first Pareto solution set of the set SI based on the set O1, wherein the decision space is composed of parameters of the biometric recognition model, selecting a second test data set S2 from the current decision space based on a Non-dominated Ranking Genetic Algorithm NSGA and obtaining a second set O2 of multi-objective values corresponding to the set S2 determined based on a trained agent model, determining a second Pareto solution set of the set S2 based on the set O2, and updating the current decision space based on the set O1, the set O2, the set S1 and the set S2 until a condition is met; and determining a final Pareto solution set based on the first Pareto solution set and the second Pareto solution set in response to the condition being satisfied.