Transfer Learning System for Deep Neural Network Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep neural networks require large amounts of data for high recognition rates, and determining domain similarity for transfer learning is time-consuming and uncertain, making it difficult to select an optimal pre-trained model for effective transfer learning.
Innovation Solution
A transfer learning system that automatically searches for an optimal pre-trained model by clustering features of transfer learning data and evaluating performance using normalized mutual information, allowing for the generation of high-performing transfer learning models even with limited data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If transfer learning is used to achieve high recognition rates with limited data, then the recognition rate is improved, but it becomes difficult to determine domain similarity and select optimal pre-trained models
Solution Approach 1:
The system performs self-service by automatically evaluating domain similarity between pre-trained models and target tasks using clustering algorithms. The pre-trained model selection unit autonomously computes similarity metrics and selects appropriate models without requiring manual expert intervention, thereby resolving the contradiction between achieving high recognition rates and reducing model selection complexity
Solution Approach 2:
The system changes parameters by computing domain similarity metrics through clustering algorithms and using these computed parameters to automatically select pre-trained models. This parameter-based automated selection process transforms the complex qualitative judgment of domain similarity into quantifiable similarity scores, enabling systematic model selection that improves recognition rates while reducing selection complexity
2Measurement precision
If manual selection of pre-trained models is performed to achieve optimal transfer learning, then the recognition rate is improved, but it requires a lot of time and expert knowledge
Solution Approach 1:
The pre-trained model selection unit implements self-service by automatically evaluating domain similarity and selecting appropriate pre-trained models through clustering algorithms. This automated process eliminates the need for manual expert intervention, significantly reducing the time required for model selection while maintaining high recognition rates through systematic similarity-based selection
Solution Approach 2:
The system replaces the manual mechanical process of expert model selection with an automated computational system. Clustering algorithms and similarity metric computations substitute for human expert analysis, transforming the time-consuming manual selection process into an efficient automated procedure that achieves optimal model selection without requiring expert knowledge or significant time investment
3Measurement precision
If various pre-training datasets and deep neural network models are used to improve transfer learning effectiveness, then the recognition rate is improved, but the complexity of selecting the optimal configuration increases
Solution Approach 1:
The pre-trained model selection unit implements a universal approach by using clustering algorithms that can evaluate domain similarity across different pre-training datasets and model types. This multi-functional evaluation framework handles diverse configurations systematically, enabling the selection of optimal pre-trained models regardless of the specific dataset or model architecture, thereby improving recognition rates while managing configuration complexity through a unified selection mechanism
Data Source
AI summary
Disclosed is a transfer learning system for a deep neural network. The transfer learning system includes a pre-trained model storage unit configured to store a plurality of pre-trained models that are deep neural network models learned using one or more pre-training datasets, a transfer learning data input unit configured to receive transfer learning data, a pre-trained model selecting unit configured to select a pre-trained model corresponding to the transfer learning data from among the plurality of stored pre-trained models, and a transfer learning unit configured to generate one or more transfer learning models by performing transfer learning using the selected pre-trained model and the transfer learning data.