Robust Loss

Robust loss functions aim to improve the performance of machine learning models by reducing sensitivity to noisy data, outliers, and adversarial attacks. Current research focuses on developing robust loss variants for various tasks, including classification, regression, time series forecasting, and reinforcement learning, often employing techniques like generative adversarial networks (GANs) and meta-learning to adapt to different noise characteristics. These advancements are significant because they enhance the reliability and generalizability of models, particularly in applications with imperfect or uncertain data, such as medical image analysis and preference learning for large language models. The resulting improvements in model robustness translate to more reliable and trustworthy predictions across diverse domains.

Papers

November 22, 2022

Improving Robust Generalization by Direct PAC-Bayesian Bound Minimization
Zifan Wang, Nan Ding, Tomer Levinboim, Xi Chen, Radu Soricut
Adversarial Robustness Robust Optimization Robust Generalization Robust Loss PAC Bayesian Bound

November 10, 2022

On Proper Learnability between Average- and Worst-case Robustness
Vinod Raman, Unique Subedi, Ambuj Tewari
Robust Loss Worst Case Robustness Average Performance Robust PAC

June 22, 2022

Optimal transport meets noisy label robust loss and MixUp regularization for domain adaptation
Kilian Fatras, Hiroki Naganuma, Ioannis Mitliagkas
Domain Adaptation Optimal Transport Domain Shift Noisy Label Training Domain Robust Loss Mixup Regularization

May 17, 2022

Robust Losses for Learning Value Functions
Andrew Patterson, Victor Liao, Martha White
Reinforcement Learning Value Function Saddle Point Gradient Update Robust Loss Policy Prediction

March 3, 2022

Semi-supervised Learning using Robust Loss
Wenhui Cui, Haleh Akrami, Anand A. Joshi, Richard M. Leahy
Semi Supervised Learning Pseudo Label Semi Supervised Training Robust Loss Robust Loss Function

Robust Loss

Papers

Improving Robust Generalization by Direct PAC-Bayesian Bound Minimization

On Proper Learnability between Average- and Worst-case Robustness

Optimal transport meets noisy label robust loss and MixUp regularization for domain adaptation

Robust Losses for Learning Value Functions

Semi-supervised Learning using Robust Loss