Noisy datasets - a xb-chang Collection

xb-chang 's Collections

Reinforcement Learning

LLMs

vision language models (VLM)

Data Generation

Noisy datasets

updated Jul 22

An accurate detection is not all you need to combat label noise in web-noisy datasets

Paper • 2407.05528 • Published Jul 8 • 3

Note 和SOTA方法(PLS, [2R])互补的noisy sample detection method, based on self-sup learning (SSL). 利用SSL学习得到的特征能较好区分ID和OOD样本，但是会把一些simple ID samples也划分为OOD，该特性和现有score based方法互补，因此交替融合，进一步提升SOTA结果。