Transfer Learning Domain Adaptation for Sparse ATE Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated testing equipment (ATE) systems face challenges in effectively training machine learning (ML) models due to the scarcity and high cost of labeled data, particularly in manufacturing environments where data collection is time-consuming and sparse, limiting the ability to accurately classify or predict product performance.
Innovation Solution
Transfer learning using Sliced Wasserstein Optimal Transport (SWOT) with smear regularization to align data distributions across domains, leveraging data from a data-rich source domain to enhance model performance in a data-sparse target domain by creating synthetic data points based on statistical moments from the source domain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional ML training is used with limited labeled data, then data collection time and cost are reduced, but model accuracy and reliability deteriorate
Solution Approach 1:
The patent creates synthetic data copies from the source domain by applying learned transport maps to transform source data into target domain representations. These synthetic copies augment the limited target domain data, enabling reliable model training without extensive time-consuming data collection. The transport map learns the statistical transformation between domains, generating realistic synthetic samples that preserve the target domain's data distribution characteristics.
2Measurement precision
If more labeled data is collected for target domain, then model accuracy improves, but data collection cost and time increase
Solution Approach 1:
The patent introduces a transport map as an intermediary that bridges the source domain (with abundant data) and the target domain (with scarce data). This intermediary learns the statistical transformation between domains and enables knowledge transfer, allowing the model to achieve high classification accuracy on target domain data without requiring large quantities of manually collected target domain samples. The intermediary effectively mediates the data scarcity problem by translating source domain knowledge into target domain representations.
3Reliability
If transfer learning is applied to align data distributions, then model performance on target domain improves, but computational complexity increases
Solution Approach 1:
The patent employs Sliced Wasserstein Optimal Transport with smear regularization, which modifies the traditional optimal transport approach by adding regularization terms that control the transport map's behavior. This parameter change in the optimization objective balances the alignment of data distributions with computational tractability. The regularization prevents overfitting and ensures the transport map remains computationally manageable while still achieving effective domain adaptation and improved model performance on the target domain.
Data Source
AI summary
System and techniques for adapting one domain to another to facilitate transfer learning in a machine learning (ML) model are described herein. Given data from a source domain and a target domain, moments can be mapped from the source domain to the target domain. Synthetic data that is constrained by the mapping can be created in the target domain. The synthetic data can be combined with the target domain data to create training data for a machine learning model to enable the model to be trained to accept target domain data as input and produce an output.