关键词: 压缩编码, 分类, 负面迁移, 迁移学习

Abstract: In real world applications such as manufacturing, a new task is often related to another existing task. Transfer learning techniques are developed to build novel models on new tasks by extracting useful information from the existing models, to reduce the high cost of inquiring the labeled information for the target task. However, how to avoid negative transfer which happens due to different distributions of tasks in a heterogeneous environment is still an open problem. Unlike traditional methods which only measure either similarity between tasks or instance relatedness, a Transfer Learning method with Compact Coding（TLCC） is proposed under a two-level framework in inductive transfer learning setting. Particularly speaking, in the macro level perspective, the degree of the similarity is represented by the relevant code length of the class boundary of each source task with respect to the target task through minimum encoding. In addition, informative instances of the source tasks are adaptively selected in the micro level viewpoint to make the choice of the specific source task more accurate. Extensive experiments show the effectiveness of the algorithm in terms of the classification accuracy in both UCI and text data sets.

Key words: compact coding, classification, negative transfer, transfer learning