Transfer Learning with Label Noise
This addresses a practical issue in transfer learning for machine learning practitioners by handling noisy labels, though it is an incremental improvement over existing methods that assume clean labels.
The paper tackles the problem of label noise in source data for transfer learning, showing that it adversely affects invariant representation learning and label shift correction, and proposes a Denoising Conditional Invariant Component (DCIC) framework that provably extracts invariant representations and estimates target label distributions without bias, with experimental verification on synthetic and real-world data.
Transfer learning aims to improve learning in target domain by borrowing knowledge from a related but different source domain. To reduce the distribution shift between source and target domains, recent methods have focused on exploring invariant representations that have similar distributions across domains. However, when learning this invariant knowledge, existing methods assume that the labels in source domain are uncontaminated, while in reality, we often have access to source data with noisy labels. In this paper, we first show how label noise adversely affect the learning of invariant representations and the correcting of label shift in various transfer learning scenarios. To reduce the adverse effects, we propose a novel Denoising Conditional Invariant Component (DCIC) framework, which provably ensures (1) extracting invariant representations given examples with noisy labels in source domain and unlabeled examples in target domain; (2) estimating the label distribution in target domain with no bias. Experimental results on both synthetic and real-world data verify the effectiveness of the proposed method.