# Evaluating the Interpretability of Sparse Autoencoders with Concept Annotations

arXiv: https://arxiv.org/abs/2606.24716 · HTML: https://mlsanity.com/paper/2606.24716/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in VLM benchmarks and visual reasoning, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 42 of its 55 references that are in the map.
- First posted on arXiv 23 Jun 2026.

## Placed in

- Multimodal LLMs / vision-language models / [VLM benchmarks and visual reasoning](https://mlsanity.com/mllm/vlm-benchmarks-and-visual-reasoning/)
- Mechanistic interpretability / [Interpreting vision and multimodal models](https://mlsanity.com/mechinterp/interpreting-vision-and-multimodal-models/)

## Builds on (most cited first, 30 of 42)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [DINOv2: Learning Robust Visual Features without Supervision](https://mlsanity.com/paper/2304.07193/) (2023)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Towards A Rigorous Science of Interpretable Machine Learning](https://mlsanity.com/paper/1702.08608/) (2017)
- [Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)](https://mlsanity.com/paper/1711.11279/) (2017)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Toy Models of Superposition](https://mlsanity.com/paper/2209.10652/) (2022)
- [Network Dissection: Quantifying Interpretability of Deep Visual Representations](https://mlsanity.com/paper/1704.05796/) (2017)
- [Scaling and evaluating sparse autoencoders](https://mlsanity.com/paper/2406.04093/) (2024)
- [Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders](https://mlsanity.com/paper/2407.14435/) (2024)
- [Linearity of Relation Decoding in Transformer Language Models](https://mlsanity.com/paper/2308.09124/) (2023)
- [Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet](https://mlsanity.com/paper/2605.29358/) (2026)
- [A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders](https://mlsanity.com/paper/2409.14507/) (2024)
- [Learning Multi-Level Features with Matryoshka Sparse Autoencoders](https://mlsanity.com/paper/2503.17547/) (2025)
- [BatchTopK Sparse Autoencoders](https://mlsanity.com/paper/2412.06410/) (2024)
- [Explainable AI (XAI): A Systematic Meta-Survey of Current Challenges and Future Opportunities](https://mlsanity.com/paper/2111.06420/) (2021)
- [Interpreting CLIP with Sparse Linear Concept Embeddings (SpLiCE)](https://mlsanity.com/paper/2402.10376/) (2024)
- [Automatically Interpreting Millions of Features in Large Language Models](https://mlsanity.com/paper/2410.13928/) (2024)
- [A Holistic Approach to Unifying Automatic Concept Extraction and Concept Importance Estimation](https://mlsanity.com/paper/2306.07304/) (2023)
- [Discover-then-Name: Task-Agnostic Concept Bottlenecks via Automated Concept Discovery](https://mlsanity.com/paper/2407.14499/) (2024)
- [Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control](https://mlsanity.com/paper/2405.08366/) (2024)
- [A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Models](https://mlsanity.com/paper/2503.05613/) (2025)
- [Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models](https://mlsanity.com/paper/2502.12892/) (2025)
- [Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models](https://mlsanity.com/paper/2504.02821/) (2025)
- [Measuring Disentanglement: A Review of Metrics](https://mlsanity.com/paper/2012.09276/) (2020)
- [Sparse autoencoders reveal selective remapping of visual concepts during adaptation](https://mlsanity.com/paper/2412.05276/) (2024)
- [Interpreting CLIP with Hierarchical Sparse Autoencoders](https://mlsanity.com/paper/2502.20578/) (2025)
- [Who's asking? User personas and the mechanics of latent misalignment](https://mlsanity.com/paper/2406.12094/) (2024)
- [Steering CLIP's vision transformer with sparse autoencoders](https://mlsanity.com/paper/2504.08729/) (2025)

## Built on by (most cited first, 2 of 2)

- [Verifying the Linear Representation Hypothesis: How Interpretable Are Vision SAEs?](https://mlsanity.com/paper/2609.35020/) (2026)
- [Towards Fast and Disentangled Counterfactuals for Visual Foundation Models](https://mlsanity.com/paper/2610.00895/) (2026)

## Abstract (arXiv)

Sparse autoencoders (SAEs) are increasingly used to extract interpretable concepts from vision and vision language models, yet existing evaluation methods largely rely on proxy metrics or qualitative inspection rather than measuring semantic correspondence. We present a human-grounded evaluation framework that quantifies alignment between SAE latents and human-annotated concepts, without requiring user studies, and validate this matching through targeted attribute perturbations. To enable this intervention-style evaluation in vision, we construct synCUB and synCOCO, synthetic benchmarks of paired images that differ in exactly one attribute. We introduce Fully-Binary Matching Pursuit (FBMP), a coalition-based matching procedure that supports many-to-one mappings between SAE latents and annotated concepts, and consistently outperforms one-to-one baselines. For functional validation, we propose a Targeted Attribute Perturbation Alignment Score (TAPAScore), which tests whether matched concepts respond selectively and in the expected direction under targeted image-level attribute perturbations. Under sanity checks, our matching and TAPAScore are the only evaluated metrics that reliably distinguish trained SAEs from untrained ones. Across SAEs trained on CLIP and DINOv2 embeddings, we find that increased overcompleteness can reduce perturbation alignment, indicating a reduction in interpretability. Our evaluation framework suggests that moderate dictionary sizes provide the best trade-off, yielding the most interpretable SAEs. Code and datasets are available at https://github.com/JonasKlotz/sae-concept-eval.