ByungKwanLee/Adversarial-Information-Bottleneck
[NeurIPS 2021] Official PyTorch Implementation for "Distilling Robust and Non-Robust Features in Adversarial Examples by Information Bottleneck"
This project helps machine learning researchers and security analysts understand how image classification models perceive visual information, especially when dealing with 'adversarial examples.' It takes images and trained classification models as input, and outputs visualizations of what parts of an image the model considers 'robust' (essential for correct classification) versus 'non-robust' (easily manipulated by adversaries). This allows users to dissect model behavior and improve its resistance to malicious inputs.
No commits in the last 6 months.
Use this if you are a machine learning researcher or security specialist interested in the internal workings of image classification models, especially concerning their vulnerability to adversarial attacks and the interpretability of features.
Not ideal if you are looking for a simple, out-of-the-box solution to directly train a robust model without diving into the underlying feature decomposition and visualization.
Stars
49
Forks
3
Language
Python
License
MIT
Category
Last pushed
Mar 13, 2023
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/ByungKwanLee/Adversarial-Information-Bottleneck"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.