# An Investigation of how Label Smoothing Affects Generalization

arXiv: https://arxiv.org/abs/2010.12648 · HTML: https://mlsanity.com/paper/2010.12648/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 23 Oct 2020.

## Built on by (most cited first, 11 of 11)

- [Are All Losses Created Equal: A Neural Collapse Perspective](https://mlsanity.com/paper/2210.02192/) (2022)
- [To Smooth or Not? When Label Smoothing Meets Noisy Labels](https://mlsanity.com/paper/2106.04149/) (2021)
- [Free Lunch for Domain Adversarial Training: Environment Label Smoothing](https://mlsanity.com/paper/2302.00194/) (2023)
- [Data Optimization in Deep Learning: A Survey](https://mlsanity.com/paper/2310.16499/) (2023)
- [The Implicit Length Bias of Label Smoothing on Beam Search Decoding](https://mlsanity.com/paper/2205.00659/) (2022)
- [Distilling Calibrated Student from an Uncalibrated Teacher](https://mlsanity.com/paper/2302.11472/) (2023)
- [Label Smoothed Embedding Hypothesis for Out-of-Distribution Detection](https://mlsanity.com/paper/2102.05131/) (2021)
- [Priority-Aware Clinical Pathology Hierarchy Training for Multiple Instance Learning](https://mlsanity.com/paper/2507.20469/) (2025)
- [Revisiting the Role of Label Smoothing in Enhanced Text Sentiment Classification](https://mlsanity.com/paper/2312.06522/) (2023)
- [Label Smoothing is a Pragmatic Information Bottleneck](https://mlsanity.com/paper/2508.14077/) (2025)
- [A Vision for Multisensory Intelligence: Sensing, Science, and Synergy](https://mlsanity.com/paper/2601.04563/) (2026)

## Abstract (arXiv)

It has been hypothesized that label smoothing can reduce overfitting and improve generalization, and current empirical evidence seems to corroborate these effects. However, there is a lack of mathematical understanding of when and why such empirical improvements occur. In this paper, as a step towards understanding why label smoothing is effective, we propose a theoretical framework to show how label smoothing provides in controlling the generalization loss. In particular, we show that this benefit can be precisely formulated and identified in the label noise setting, where the training is partially mislabeled. Our theory also predicts the existence of an optimal label smoothing point, a single value for the label smoothing hyperparameter that minimizes generalization loss. Extensive experiments are done to confirm the predictions of our theory. We believe that our findings will help both theoreticians and practitioners understand label smoothing, and better apply them to real-world datasets.