计算机科学
域适应
分类器(UML)
学习迁移
子空间拓扑
人工智能
相似性(几何)
模式识别(心理学)
源代码
算法
机器学习
相关性
数学
几何学
图像(数学)
操作系统
作者
Weiyou Yang,Minghua Wang,Xiaoke Ma
标识
DOI:10.1109/bibm58861.2023.10385717
摘要
Multi-source transfer learning has received widespread concern due to outstanding predictive performance. But the limitations imposed by the initial outputs of the source classifiers on the decision values of the target samples, as well as the significant differences between the source and target domains, negative transfer may occur. A novel algorithm called Multi-Source Adaption and Similarity Learning (MSASL) is proposed, which utilizes the predicted probabilities of each target source classifier and the learned correlation probabilities to calculate the classification probabilities of target domain samples. It can dynamically learn the correlations between different source and target domains in a common subspace and discard some irrelevant source domains by measuring the domain differences. Simulation experiments have demonstrated the effectiveness of the method.
科研通智能强力驱动
Strongly Powered by AbleSci AI