Learning with Instance-Dependent Label Noise: A Sample Sieve Approach

Authors: Hao Cheng, Zhaowei Zhu, Xingyu Li, Yifei Gong, Xing Sun, Yang Liu

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the performance of CORES2 on CIFAR10 and CIFAR100 datasets with synthetic instance-dependent label noise and Clothing1M with real-world human noise. ... CORES2 achieves competitive performance on multiple datasets, including CIFAR-10, CIFAR-100, and Clothing1M, under different label noise settings.
Researcher Affiliation Collaboration University of California, Santa Cruz, Tencent You Tu Lab {zwzhu,xli279,yangliu}@ucsc.edu, {louischeng,yifeigong,winfredsun}@tencent.com
Pseudocode Yes Algorithm 1 Instance-Dependent Label Noise Generation
Open Source Code Yes Code is available at https://github.com/UCSC-REAL/cores.
Open Datasets Yes Datasets: CORES2 is evaluated on three benchmark datasets: CIFAR-10, CIFAR-100 (Krizhevsky et al., 2009) and Clothing1M (Xiao et al., 2015).
Dataset Splits Yes Standard data augmentation is applied to each dataset. ... We use Res Net34 for CIFAR-10 and CIFAR-100 and Res Net50 for Clothing1M.
Hardware Specification No Only mentions using Res Net34 for CIFAR-10 and CIFAR-100 and Res Net50 for Clothing1M, which are model architectures, not hardware specifications for running experiments. No details on GPUs, CPUs, or memory were provided.
Software Dependencies No The paper mentions optimizer (SGD) and loss functions (CE, KL-divergence) but does not provide specific software library names with version numbers (e.g., PyTorch 1.9, TensorFlow 2.x).
Experiment Setup Yes The basic hyper-parameters settings for CIFAR-10 and CIFAR-100 are listed as follows: mini-batch size (64), optimizer (SGD), initial learning rate (0.1), momentum (0.9), weight decay (0.0005), number of epochs (100) and learning rate decay (0.1 at 50 epochs).