Sample data intelligent screening enhancement processing method for artificial intelligence model training
By employing a closed-loop framework of dynamic adaptive filtering and knowledge graph-constrained generative enhancement, the disconnect between filtering and enhancement in sample data processing is resolved, thereby improving sample quality and optimizing model performance. This framework adapts to dynamic changes in model training and enhances processing efficiency and accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- MAGIC BRUSH MA LIANG ARTIFICIAL INTELLIGENCE (HANGZHOU) CO LTD
- Filing Date
- 2026-01-27
- Publication Date
- 2026-05-26
AI Technical Summary
In existing technologies, the sample data processing suffers from several problems: the screening process relies on static rules, leading to the omission of core difficult samples or an excessive number of redundant samples; the enhancement strategies lack domain knowledge constraints, resulting in sample distortion and semantic shift; the screening and enhancement processes lack synergy, resulting in low processing efficiency and difficulty in meeting the training requirements of high-precision models.
A dynamic adaptive screening module is used to select samples by combining static quality indicators and dynamic training feedback indicators. Enhanced samples are generated through a knowledge graph-constrained generative enhancement module. The strategy is optimized through a screening and enhancement linkage control module, thus constructing a closed-loop framework for sample processing and model training to ensure the quality of enhanced samples and adapt to dynamic changes in the model.
It significantly improved sample quality and processing efficiency, optimized model accuracy and generalization ability, shortened training convergence time, and improved model performance.
Smart Images

Figure CN122090199A_ABST