Learning from Similarity-Confidence Data

Authors: Yuzhou Cao, Lei Feng, Yitian Xu, Bo An, Gang Niu, Masashi Sugiyama

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on various datasets and deep neural networks clearly demonstrate the effectiveness of the proposed Sconf learning method and risk correction scheme (in Section 7).
Researcher Affiliation Academia 1College of Science, China Agricultural University, Beijing, China 2College of Computer Science, Chongqing University, Chongqing, China 3RIKEN Center for Advanced Intelligence Project, Tokyo, Japan 4Nanyang Technological University, School of Computer Science and Engineering, Singapore 5The University of Tokyo, Tokyo, Japan.
Pseudocode No The paper does not include a section or figure explicitly labeled "Pseudocode" or "Algorithm" with structured steps.
Open Source Code No The paper does not provide an explicit statement about open-sourcing its code or a link to a code repository.
Open Datasets Yes We evaluated the performance of proposed methods on six widely-used benchmarks MNIST (Le Cun et al., 1998), Fashion-MNIST (Xiao et al.), Kuzushiji-MNIST (Clanuwat et al., 2018), EMNIST (Cohen et al., 2017), SVHN (Netzer et al., 2011), and CIFAR-10 (Krizhevsky, 2012).
Dataset Splits No For ERM-based methods: Sconf-Unbiased, Sconf-ABS, Sconf-NN, and SD, the validation accuracy was also calculated according to their empirical risk estimators on a validation set consisted of Sconf data, which means that we do not have to collect additional ordinarily labeled data for validation when using ERM-based methods.
Hardware Specification Yes We implemented all the methods by Pytorch (Paszke et al., 2019), and conducted the experiments on NVIDIA Tesla P4 GPUs.
Software Dependencies No The paper mentions "Pytorch (Paszke et al., 2019)" and "Adam (Kingma & Ba, 2015)" but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes We trained the model with Adam for 100 epochs (full-batch size) and default momentum parameter. The learning rate was initially set to 0.1 and divided by 10 every 30 epochs.