Masking: A New Perspective of Noisy Supervision
Authors: Bo Han, Jiangchao Yao, Gang Niu, Mingyuan Zhou, Ivor Tsang, Ya Zhang, Masashi Sugiyama
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on CIFAR-10 and CIFAR-100 with three noise structures as well as the industrial-level Clothing1M with agnostic noise structure, and the results show that Masking can improve the robustness of classifiers significantly. |
| Researcher Affiliation | Academia | 1Centre for Artificial Intelligence, University of Technology Sydney 2Center for Advanced Intelligence Project, RIKEN 3Cooperative Medianet Innovation Center, Shanghai Jiao Tong University 4Mc Combs School of Business, The University of Texas at Austin 5Graduate School of Frontier Sciences, University of Tokyo |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The implementation is available at https://github.com/bhan ML/Masking. |
| Open Datasets | Yes | CIFAR-10 and CIFAR-100 datasets are used. Both datasets consist of 50k samples for training and 10k samples for testing, where each sample is a 32 32 color image and its label. For CIFAR-10, we randomly flip the labels of the training set according to the first two types of noise structure... An industrial-level dataset called Clothing1M [46] from online shopping websites (i.e., Taobao.com) is used here, where the ground-truth transition matrix is not available. |
| Dataset Splits | No | The paper specifies training and testing splits, but does not provide explicit details about a separate validation dataset split used in their experiments. It mentions the concept of validation sets in a discussion but not as part of their concrete experimental setup for data partitioning. |
| Hardware Specification | Yes | All experiments are conducted on a NVIDIA TITAN GPU... We gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan Xp GPU used for this research. |
| Software Dependencies | No | The paper states that methods are "implemented by Tensorflow" but does not provide specific version numbers for TensorFlow or any other software dependencies. |
| Experiment Setup | Yes | For both datasets, the batch size is set to 128 for 15,000 iterations. α and β in Eq. (2) are respectively set 0.05 and 0.005. |