Are Anchor Points Really Indispensable in Label-Noise Learning?
Authors: Xiaobo Xia, Tongliang Liu, Nannan Wang, Bo Han, Chen Gong, Gang Niu, Masashi Sugiyama
NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results on benchmark-simulated and real-world label-noise datasets demonstrate that without using exact anchor points, the proposed method is superior to state-of-the-art label-noise learning methods. |
| Researcher Affiliation | Academia | 1University of Sydney 2Xidian University 3RIKEN 4Nanjing University of Science and Technology 5University of Tokyo |
| Pseudocode | Yes | Algorithm 1 Reweight T-Revision (Reweight-R) Algorithm. |
| Open Source Code | No | The paper does not provide an explicit statement about open-source code availability or a link to a code repository for the methodology described. |
| Open Datasets | Yes | Datasets We verify the effectiveness of the proposed method on three synthetic noisy datasets, i.e., MNIST [19], CIFAR-10 [18], and CIFAR-100 [18], and one real-world noisy dataset, i.e., clothing1M [44]. |
| Dataset Splits | Yes | For all the datasets, we leave out 10% of the training examples as a validation set. The three datasets contain clean data. We corrupted the training and validation sets manually according to true transition matrices T. ... MNIST has 10 classes of images including 60,000 training images and 10,000 test images. CIFAR-10 has 10 classes of images including 50,000 training images and 10,000 test images. CIFAR100 also has 50,000 training images and 10,000 test images, but 100 classes. |
| Hardware Specification | Yes | For fair comparison, we implement all methods with default parameters by Py Torch on NVIDIA Tesla V100. |
| Software Dependencies | No | The paper mentions implementing methods "by Py Torch" but does not specify a version number for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | We use a Le Net-5 network for MNIST, a Res Net-18 network for CIFAR-10, a Res Net-34 network for CIFAR-100. For learning the transition matrix ˆT in the first stage, we follow the optimization method in [30]. During the second stage, we first use SGD with momentum 0.9, weight decay 10 4, batch size 128, and an initial learning rate of 10 2 to initialize the network. The learning rate is divided by 10 after the 40th epoch and 80th epoch. 200 epochs are set in total. Then, the optimizer and learning rate are changed to Adam and 5 10 7 to learn the classifier and slack variable. For CIFAR-10 and CIFAR-100, we perform data augmentation by horizontal random flips and 32 32 random crops after padding 4 pixels on each side. For clothing1M, we use a Res Net-50 pre-trained on Image Net. Follow [30], we also exploit the 1M noisy data and 50k clean data to initialize the transition matrix. In the second stage, for initialization, we use SGD with momentum 0.9, weight decay 10 3, batch size 32, and run with learning rates 10 3 and 10 4 for 5 epochs each. For learning the classifier and slack variable, Adam is used and the learning rate is changed to 5 10 7. |