Open-set Label Noise Can Improve Robustness Against Inherent Label Noise
Authors: Hongxin Wei, Lue Tao, RENCHUNZI XIE, Bo An
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results on benchmark datasets with various types of noisy labels demonstrate that the proposed method not only enhances the performance of many existing robust algorithms but also achieves significant improvement on Out-of-Distribution detection tasks even in the label noise setting. |
| Researcher Affiliation | Academia | Hongxin Wei1 Lue Tao2,3 Renchunzi Xie1 Bo An1 1School of Computer Science and Engineering, Nanyang Technological University, Singapore 2 College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, China 3 MIIT key Laboratory of Pattern Analysis and Machine Intelligence, China |
| Pseudocode | Yes | Algorithm 1 Open-set Regularization with Dynamic Noisy Labels |
| Open Source Code | Yes | The code is published on https://github.com/hongxin001/ODNL |
| Open Datasets | Yes | We conduct extensive experiments on both simulated and real-world noisy datasets, including CIFAR-10, CIFAR-100 and Clothing1M datasets. |
| Dataset Splits | Yes | We use 5k noisy samples as the validation to tune the hyperparameter η in {0.1, 0.5, 1, 2.5, 5}, then train the model on the full training set and report the average test accuracy in the last 5 epochs. |
| Hardware Specification | No | The paper states 'See Appendix B' for details on 'the total amount of compute and the type of resources used', but Appendix B is not provided in the current document, thus specific hardware details are not available in the given text. |
| Software Dependencies | No | The paper does not explicitly provide specific software dependencies with version numbers (e.g., library or solver names with version numbers) needed to replicate the experiment. |
| Experiment Setup | Yes | The network is trained for 200 epochs using SGD with a momentum of 0.9, a weight decay of 0.0005. In each iteration, both the batch sizes of the original dataset and the open-set auxiliary dataset are set as 128. We set the initial learning rate as 0.1, and reduce it by a factor of 10 after 80 and 140 epochs. |