Label Noise: Ignorance Is Bliss

Authors: Yilun Zhu, Jianxin Zhang, Aditya Gangrade, Clay Scott

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We establish a new theoretical framework for learning under multi-class, instancedependent label noise. ... Finally, we translate this theoretical insight into practice: by using NI-ERM to fit a linear classifier on top of a self-supervised feature extractor, we achieve state-of-the-art performance on the CIFAR-N data challenge. ... We conducted experiments 1 on the CIFAR image data under two scenarios: synthetic label flipping (symmetric noise) and realistic human label errors [Wei et al., 2022], as shown in Figure 3.
Researcher Affiliation Academia Yilun Zhu EECS University of Michigan allanzhu@umich.edu Jianxin Zhang EECS University of Michigan jianxinz@umich.edu Aditya Gangrade ECE Boston University gangrade@bu.edu Clayton Scott EECS, Statistics University of Michigan clayscot@umich.edu
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Code is available at: https://github.com/allan-z/label_noise_ignorance. ... Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: Code is provided, common benchmark datase were used, instructions are given, the result is easily reproducible.
Open Datasets Yes CIFAR-N data challenge. ... We conducted experiments 1 on the CIFAR image data under two scenarios: synthetic label flipping (symmetric noise) and realistic human label errors [Wei et al., 2022], as shown in Figure 3. ... MNIST (http://yann.lecun.com/exdb/mnist/) ... CIFAR-10 (https://www.cs.toronto.edu/~kriz/cifar. html)
Dataset Splits Yes We prespecify a range of values for ℓ2 regularization ({0.0001, 0.001, 0.01, 0.1, 1, 10, 100} ) and number of iterations for lbfgs optimizer ({10, 20, 50, 100}), then do cross-validation on noisy data to pick the best hyper-parameters.
Hardware Specification Yes The experiment was conducted on AMD Ryzen 5 3600 CPU. ... The experiments were conducted on a single NVIDIA GTX 1660S GPU. ... The experiments were conducted on a single NVIDIA Tesla V100 GPU.
Software Dependencies No The paper mentions software components like "Sklearn's logistic regression" and "Pytorch model library" but does not provide specific version numbers for these software dependencies, which are required for full reproducibility.
Experiment Setup Yes We prespecify a range of values for ℓ2 regularization ({0.0001, 0.001, 0.01, 0.1, 1, 10, 100} ) and number of iterations for lbfgs optimizer ({10, 20, 50, 100}), then do cross-validation on noisy data to pick the best hyper-parameters.