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

arXiv:2404.02949v1 Announce Type: new
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 …

abstract ai systems arxiv cnn cnns competition concept convolutional neural networks cs.ai cs.lg human humans identify imagenet innovations interpretability networks neural networks novel scale studying systems type workers