Multi-Dataset Pre-Trained Model Selection for Generalizable Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing model training methods consider only matching degrees between pre-trained models and user tasks, resulting in suboptimal final models with poor user experience.
Innovation Solution
Evaluate pre-trained models based on multiple datasets to determine a first and second model, then train a target model using both to enhance generalization capability and function.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If only matching degree between pre-trained models and user tasks is considered, then model selection is simple, but final model performance and generalization capability are poor
Solution Approach 1:
The patent changes the evaluation parameters from single matching degree to multiple dimensions including performance metrics (accuracy, precision, recall, F1-score) and generalization capability metrics. This multi-parameter evaluation system resolves the contradiction by providing comprehensive model assessment without significantly increasing operational complexity.
Solution Approach 2:
The patent creates a universal evaluation framework that can assess pre-trained models across multiple dimensions and tasks simultaneously. The evaluation module performs multiple functions: performance evaluation, generalization assessment, and model ranking, thereby improving reliability while maintaining operational efficiency through a consolidated evaluation process.
2Reliability
If multiple evaluation dimensions are used to assess pre-trained models, then final model quality improves, but evaluation complexity increases
Solution Approach 1:
The patent segments the evaluation process into distinct modules: performance evaluation module and generalization capability evaluation module. Each module handles specific evaluation dimensions independently, computing metrics like accuracy, precision, recall for performance, and cross-dataset validation for generalization. This segmentation reduces overall complexity by breaking down the complex evaluation task into manageable, specialized components.
Solution Approach 2:
The patent introduces an evaluation module as an intermediary between pre-trained models and final model selection. This intermediary performs comprehensive multi-dimensional assessment and translates complex evaluation results into actionable model rankings and selections, thereby managing evaluation complexity while maintaining high evaluation accuracy through a dedicated intermediary system.
3Reliability
If pre-trained models are selected based on comprehensive evaluation, then user experience improves, but training time and computational resources increase
Solution Approach 1:
The patent performs preliminary evaluation of pre-trained models before final model training. The evaluation module assesses performance and generalization capabilities of candidate pre-trained models in advance, identifying the most suitable models for further training. This preliminary action filters out inferior models early, reducing the number of models requiring extensive training and thereby reducing overall training time while maintaining high user experience quality.
Solution Approach 2:
The patent changes training parameters dynamically based on evaluation results. Models with higher evaluation scores receive more training resources, while lower-scoring models receive reduced training or are discarded. This parameter adjustment optimizes the trade-off between training time and final model quality, ensuring that computational resources are allocated efficiently to models most likely to deliver good user experience.
Data Source
AI summary
The present disclosure relates to model training methods and devices. One example method includes obtaining a plurality of datasets of a target task, evaluating a plurality of pre-trained models, based on the plurality of datasets, to obtain evaluation values of the plurality of pre-trained models, where the evaluation values indicate differences between performance of the pre-trained models on the plurality of datasets, determining a first pre-trained model and a second pre-trained model from the plurality of pre-trained models, where the first pre-trained model matches the target task relatively best among the plurality of pre-trained models, and the second pre-trained model has a relatively highest evaluation value or a relatively lowest evaluation value among the plurality of pre-trained models, and training a to-be-trained model, based on the plurality of datasets, the first pre-trained model, and the second pre-trained model, to obtain a target model.