AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Evaluating the Interpretability of Sparse Autoencoders with Concept Annotations

Jonas Klotz, Cassio F. Dantas, Pallavi Jain, Diego Marcos, Begüm Demir

ECCV 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Language-grounded visual concept learning · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use many-to-one FBMP coalitions (k=3) instead of one-to-one F1 matching (ablation)SAEs (BatchTopK, Matryoshka, TopK, JumpReLU) on CLIP · CUB and COCO matchingDelta MATCHScoreImproved
use F_0.5 instead of F_1 or F_0.25 as the FBMP selection criterion (ablation)FBMP · CUB and COCO matchingDelta MATCHScoreImproved
increase dictionary size (ablation)TopK SAE (CLIP, CUB) · synCUBTAPAScoreHurt
increase dictionary size beyond 1024 (ablation)BatchTopK, TopK, JumpReLU SAEs (CLIP, COCO) · synCOCOTAPAScoreHurt
increase dictionary size (ablation)BatchTopK and Matryoshka SAEs (CLIP, CUB) · CUBDelta MATCHScoreMixed
increase dictionary size to 4096 (ablation)TopK SAE (CLIP, CUB) · CUBDelta MATCHScoreMixed
use trained SAE instead of untrained or random activations, measured with FBMP matching and TAPAScoresanity check on CUB and COCO · CLIP embeddingsmetric separationImproved
compare existing proxy metrics (MS, FMS, CKNNA) across trained vs untrained SAEMS, FMS, CKNNA · CUB and COCO sanity checkmetric sensitivityNo clear effect
matching score as predictor of TAPAScore (correlation)FBMP F0.5 · synCUBPearson correlationImproved
matching score as predictor of TAPAScore (correlation)FBMP and F1 k=1 · synCOCOPearson correlationHurt
use Matryoshka nested dictionaries (ablation)SAE variant · synCOCOTAPAScoreMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy