Skip to content

Opening book details…

About this document

Label Smoothing and Label Noise Mitigation by harsha vardhan is a document available to read on EtoBox.

This paper investigates the effectiveness of label smoothing as a technique for mitigating label noise in deep learning models. It finds that while label smoothing can amplify label noise, it is also competitive with loss-correction techniques and improves model performance under noisy conditions. The authors demonstrate that label smoothing enhances the accuracy of model distillation from noisy data, suggesting its viability as a denoising method.

Author
harsha vardhan
Language
EN