Optimization method and system for searching combined loss function based on genetic algorithm

By constructing a finite search space and combining outer and inner layer optimization, the problem of unstable training of the combined loss function is solved by using a genetic algorithm-based combined loss function optimization method. This improves the stability and generalization ability of the model and reduces the cost of manual parameter tuning.

CN121834359BActive Publication Date: 2026-06-12LIMING VOCATIONAL UNIV +1
View PDF 4 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
LIMING VOCATIONAL UNIV
Filing Date
2026-03-13
Publication Date
2026-06-12

AI Technical Summary

⚠Technical Problem

Existing models suffer from problems such as parameter dependence on manual adjustment, unstable training process, and high dimensionality of search space when training the combined loss function under conditions of small sample size, imbalanced positive and negative samples, and imbalanced easy and difficult samples.

⚗Method used

A genetic algorithm-based search optimization method for the combined loss function is adopted. By applying non-negativity constraints, normalization constraints, and interval pruning to the combined loss parameter vector, a finite feasible search space is constructed. The weights and internal parameters of the combined loss function are optimized by combining the outer genetic search and the inner fixed-budget gradient optimization.

🎯Benefits of technology

It improves the stability and generalization performance of model training, reduces the cost of manual parameter tuning, is suitable for searching non-differentiable and discretized loss configurations, and improves the training stability and reproducibility under small sample data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121834359B_ABST
    Figure CN121834359B_ABST
Patent Text Reader

Abstract

The application discloses a combination loss function search optimization method and system based on a genetic algorithm, and belongs to the technical field of computer vision. The search optimization method comprises the following steps: constructing a combination loss function L , forming a combination loss parameter vector; applying non-negative constraints, normalization constraints and interval clipping constraints to map the combination loss parameter into a finite discrete feasible search space; performing outer layer search by using a genetic algorithm, and ensuring that the optimal fitness in each generation population is not worse than that of the last generation; performing preset step number training on a predetermined model by using a gradient optimization method in an inner layer, and calculating a performance index as fitness; repeating the above steps until a preset termination condition is met, and obtaining an optimal combination loss parameter vector; and performing complete training on the predetermined model. The system comprises a parameter coding module, a constraint mapping module, an outer layer search module, an inner layer training and evaluation module and an output module. The application can improve the training stability and generalization performance of a model, and reduce the artificial parameter adjustment cost.
Need to check novelty before this filing date? Find Prior Art