Pytorch implementation of "Peer Loss Functions: Learning from Noisy Labels without Knowing Noise Rates"

Kushal Shingote

Last update: Feb 8, 2022

Related tags

Miscellaneous Multi-class-Peer-Loss-functions

Overview

Peer Loss functions

This repository is the (Multi-Class & Deep Learning) Pytorch implementation of "Peer Loss Functions: Learning from Noisy Labels without Knowing Noise Rates" accepted by ICML2020.

Required Packages & Environment

Supported OS: Windows, Linux, Mac OS X; Python: 3.6/3.7;

Deep Learning Library: PyTorch (GPU required)

Required Packages: Numpy, Pandas, random, sklearn, tqdm, csv, torch (Keras is required if you want to estimate the noise transition matrix).

Utilities

This repository includes:

📋 Multi-class implementation of Peer Loss functions;

📋 Peer Loss functions in Deep Learning;

📋 Dynamical tunning strategies of Peer Loss functions to further improve the performance.

Details of running (weighted) Peer Loss functions on MNIST, Fashion MNIST, CIFAR-10, CIFAR-100 with different noise setting are mentioned in the README.md file in each folder.

The workflow of weighted Peer Loss functions comes to:

Decision boundary visualization

Given a 2D syntheric dataset, the decision boundaries returned by training with Cross-entropy loss become loose when the noise rate is high. However, the decision boundaries w.r.t. Peer Loss functions remain tight despite high presence of label noise.

Citation

If you use our code, please cite the following paper:

@inproceedings{liu2020peer,
  title={Peer loss functions: Learning from noisy labels without knowing noise rates},
  author={Liu, Yang and Guo, Hongyi},
  booktitle={International Conference on Machine Learning},
  pages={6226--6236},
  year={2020},
  organization={PMLR}
}

Related Code

📋 Peer Loss functions and its experiments on UCI datasets is available at: https://github.com/gohsyi/PeerLoss

📋 The following work extends Peer Loss functions to a family of divergence-type robust loss functions and is available at: https://github.com/weijiaheng/Robust-f-divergence-measures The corresponding paper is "When Optimizing f-Divergence is Robust with Label noise" accepted by ICLR2021.

Thanks for watching!

You might also like...

fast_bss_eval is a fast implementation of the bss_eval metrics for the evaluation of blind source separation.

Python implementation of an automatic parallel parking system in a virtual environment, including path planning, path tracking, and parallel parking

Automatic Parallel Parking: Path Planning, Path Tracking & Control This repository contains a python implementation of an automatic parallel parking s

134 Jan 9, 2023

Comments

CVE-2007-4559 Patch

Patching CVE-2007-4559

Hi, we are security researchers from the Advanced Research Center at Trellix. We have began a campaign to patch a widespread bug named CVE-2007-4559. CVE-2007-4559 is a 15 year old bug in the Python tarfile package. By using extract() or extractall() on a tarfile object without sanitizing input, a maliciously crafted .tar file could perform a directory path traversal attack. We found at least one unsantized extractall() in your codebase and are providing a patch for you via pull request. The patch essentially checks to see if all tarfile members will be extracted safely and throws an exception otherwise. We encourage you to use this patch or your own solution to secure against CVE-2007-4559. Further technical information about the vulnerability can be found in this blog.

If you have further questions you may contact us through this projects lead researcher Kasimir Schulz.

opened by TrellixVulnTeam 0