Noise Robust Loss

Noise-robust loss functions aim to improve the performance of machine learning models trained on datasets containing noisy labels or corrupted data, a common problem hindering accurate model training. Current research focuses on developing loss functions that adaptively adjust their sensitivity to noise based on data quality or instance-specific characteristics, often employing techniques like logit bias adjustments or meta-learning to optimize hyperparameters. These advancements are crucial for enhancing the reliability and generalizability of models across various applications, particularly in domains with inherently noisy data like image retrieval, medical image analysis, and natural language processing. The ultimate goal is to create robust models that are less susceptible to errors introduced by imperfect data.

Papers

May 28, 2024

AdapNet: Adaptive Noise-Based Network for Low-Quality Image Retrieval
Sihe Zhang, Qingdong He, Jinlong Peng, Yuxi Li, Zhengkai Jiang, Jiafu Wu, Mingmin Chi, Yabiao Wang, Chengjie Wang
Network Programming Adaptive Importance Image Quality Image Retrieval High Quality Noise Robust Loss

June 8, 2023

Noise-Robust Loss Functions: Enhancing Bounded Losses for Large-Scale Noisy Data Learning
Max Staats, Matthias Thamm, Bernd Rosenow
Training Data Loss Function Model Overfitting Robust Loss Function Noise Robust Loss

January 18, 2023

Improve Noise Tolerance of Robust Loss via Noise-Awareness
Kehui Ding, Jun Shu, Deyu Meng, Zongben Xu
Robust Loss Noise Tolerance Noise Robust Loss

December 31, 2022

Tracing the Origin of Adversarial Attack for Forensic Investigation and Deterrence
Han Fang, Jiyi Zhang, Yupeng Qiu, Ke Xu, Chengfang Fang, Ee-Chien Chang
Adversarial Attack Adversarial Example Visual Creation Forensic Analysis Noise Robust Loss

September 16, 2022

Weakly Supervised Medical Image Segmentation With Soft Labels and Noise Robust Loss
Banafshe Felfeliyan, Abhilash Hareendranathan, Gregor Kuntze, Stephanie Wichuk, Nils D. Forkert, Jacob L. Jaremko, Janet L. Ronsky
Segmentation Model Soft Label Expert Annotation Noise Robust Loss

August 5, 2022

A Noise-Robust Loss for Unlabeled Entity Problem in Named Entity Recognition
Wentao Kang, Guijun Zhang, Xiao Fu
Natural Language Processing Entity Recognition Named Entity Recognition Cross Entropy Large Scale Annotated Noise Robust Loss

April 18, 2022

Adaptive Noisy Data Augmentation for Regularized Estimation and Inference in Generalized Linear Models
Yinan Li, Fang Liu
Scientific Inference Generalized Linear Model Regularization Effect Group Lasso Adaptive Lasso Noise Robust Loss

March 3, 2022

On Learning Contrastive Representations for Learning with Noisy Labels
Li Yi, Sheng Liu, Qi She, A. Ian McLeod, Boyu Wang
Noisy Label Cross Entropy Loss Contrastive Representation Learning Contrastive Representation Softmax Cross Entropy Noise Robust Loss

December 8, 2021

The perils of being unhinged: On the accuracy of classifiers minimizing a noise-robust convex loss
Philip M. Long, Rocco A. Servedio
Loss Function Label Noise Simple Classifier Binary Classifier Convex Loss Function Noise Robust Loss

Noise Robust Loss

Papers

AdapNet: Adaptive Noise-Based Network for Low-Quality Image Retrieval

Noise-Robust Loss Functions: Enhancing Bounded Losses for Large-Scale Noisy Data Learning

Improve Noise Tolerance of Robust Loss via Noise-Awareness

Tracing the Origin of Adversarial Attack for Forensic Investigation and Deterrence

Weakly Supervised Medical Image Segmentation With Soft Labels and Noise Robust Loss

A Noise-Robust Loss for Unlabeled Entity Problem in Named Entity Recognition

Adaptive Noisy Data Augmentation for Regularized Estimation and Inference in Generalized Linear Models

On Learning Contrastive Representations for Learning with Noisy Labels

The perils of being unhinged: On the accuracy of classifiers minimizing a noise-robust convex loss