The paper explores properties of Concept Activation Vectors (CAVs) in deep learning models, highlighting inconsistencies, entanglement, and spatial dependence. Tools are introduced to detect and mitigate these properties for improved interpretability.

https://arxiv.org/abs//2404.03713

YouTube: https://www.youtube.com/@ArxivPapers

TikTok: https://www.tiktok.com/@arxiv_papers

Apple Podcasts: https://podcasts.apple.com/us/podcast/arxiv-papers/id1692476016

Spotify: https://podcasters.spotify.com/pod/show/arxiv-papers