$\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise

Jialiang Wang, Xiong Zhou, Deming Zhai, Junjun Jiang, Xiangyang Ji, Xianming Liu

Keywords: Learning with Noisy Labels, Robust Loss Function, Excess Risk Bound

TL;DR: We propose a simple yet effective method for mitigating label noise, which can be implemented with just two lines of code.

Abstract: Noisy labels pose a common challenge for training accurate deep neural networks. To mitigate label noise, prior studies have proposed various robust loss functions to achieve noise tolerance in the presence of label noise, particularly symmetric losses. However, they usually suffer from the underfitting issue due to the overly strict symmetric condition. In this work, we propose a simple yet effective approach for relaxing the symmetric condition, namely **$\epsilon$-softmax**, which simply modifies the outputs of the softmax layer to approximate one-hot vectors with a controllable error $\epsilon$. Essentially, ***$\epsilon$-softmax** not only acts as an alternative for the softmax layer, but also implicitly plays the crucial role in modifying the loss function.* We prove theoretically that **$\epsilon$-softmax** can achieve noise-tolerant learning with controllable excess risk bound for almost any loss function. Recognizing that **$\epsilon$-softmax**-enhanced losses may slightly reduce fitting ability on clean datasets, we further incorporate them with one symmetric loss, thereby achieving a better trade-off between robustness and effective learning. Extensive experiments demonstrate the superiority of our method in mitigating synthetic and real-world label noise.

Primary Area: Optimization for deep networks

Submission Number: 1418