English

The SaTML '24 CNN Interpretability Competition: New Innovations for Concept-Level Interpretability

Machine Learning 2024-04-05 v1 Artificial Intelligence

Abstract

Interpretability techniques are valuable for helping humans understand and oversee AI systems. The SaTML 2024 CNN Interpretability Competition solicited novel methods for studying convolutional neural networks (CNNs) at the ImageNet scale. The objective of the competition was to help human crowd-workers identify trojans in CNNs. This report showcases the methods and results of four featured competition entries. It remains challenging to help humans reliably diagnose trojans via interpretability tools. However, the competition's entries have contributed new techniques and set a new record on the benchmark from Casper et al., 2023.

Keywords

Cite

@article{arxiv.2404.02949,
  title  = {The SaTML '24 CNN Interpretability Competition: New Innovations for Concept-Level Interpretability},
  author = {Stephen Casper and Jieun Yun and Joonhyuk Baek and Yeseong Jung and Minhwan Kim and Kiwan Kwon and Saerom Park and Hayden Moore and David Shriver and Marissa Connor and Keltin Grimes and Angus Nicolson and Arush Tagade and Jessica Rumbelow and Hieu Minh Nguyen and Dylan Hadfield-Menell},
  journal= {arXiv preprint arXiv:2404.02949},
  year   = {2024}
}

Comments

Competition for SaTML 2024

R2 v1 2026-06-28T15:43:20.978Z