Information
- Title: Mitigating Neural Network Overconfidence with Logit Normalization
- Author: Hongxin Wei, Renchunzi Xie, Hao Cheng, Lei Feng, Bo An, Yixuan Li
- Institution: 新加坡南洋理工大学,南京大学,重庆大学
- Year: 2022
- Journal: ICML
- Source: Arxiv, GitHub
- Cite: Hongxin Wei, Renchunzi Xie, Hao Cheng, Lei Feng, Bo An, Yixuan Li Proceedings of the 39th International Conference on Machine Learning, PMLR 162:23631-23644, 2022.
- Idea: 提出了通过归一化 logit 的范数来使模型给出更保守的预测,使得 ID 数据更具有可分性。