Transfer Learning with Label Noise

TL;DR

This paper addresses the challenge of transfer learning with noisy source labels by proposing a denoising framework that maintains invariant representations and accurately estimates target label distribution, improving transfer learning robustness.

Contribution

It introduces the Denoising Conditional Invariant Component (DCIC) framework, which effectively handles label noise in transfer learning scenarios, ensuring invariant representation learning and unbiased label distribution estimation.

Findings

DCIC improves invariant representation learning with noisy labels.

The method accurately estimates target label distribution.

Experimental results validate effectiveness on synthetic and real data.

Abstract

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.