# Interpretable and Testable Vision Features via Sparse Autoencoders

arXiv: https://arxiv.org/abs/2502.06755 · HTML: https://mlsanity.com/paper/2502.06755/

## Key facts

- Cited by 25 papers in the map (data through 14 Oct 2026).
- Placed in Interpreting vision and multimodal models, one of 13 directions in Mechanistic interpretability.
- Builds on 44 of its 67 references that are in the map.
- First posted on arXiv 10 Feb 2025.

## Placed in

- Mechanistic interpretability / [Interpreting vision and multimodal models](https://mlsanity.com/mechinterp/interpreting-vision-and-multimodal-models/)

## Builds on (most cited first, 30 of 44)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [ImageNet Large Scale Visual Recognition Challenge](https://mlsanity.com/paper/1409.0575/) (2014)
- [DINOv2: Learning Robust Visual Features without Supervision](https://mlsanity.com/paper/2304.07193/) (2023)
- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization](https://mlsanity.com/paper/1610.02391/) (2016)
- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [Distributed Representations of Words and Phrases and their Compositionality](https://mlsanity.com/paper/1310.4546/) (2013)
- [Axiomatic Attribution for Deep Networks](https://mlsanity.com/paper/1703.01365/) (2017)
- [Learning Deep Features for Discriminative Localization](https://mlsanity.com/paper/1512.04150/) (2015)
- [Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps](https://mlsanity.com/paper/1312.6034/) (2013)
- [Visualizing and Understanding Convolutional Networks](https://mlsanity.com/paper/1311.2901/) (2013)
- [Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)](https://mlsanity.com/paper/1711.11279/) (2017)
- [Sanity Checks for Saliency Maps](https://mlsanity.com/paper/1810.03292/) (2018)
- [Concept Bottleneck Models](https://mlsanity.com/paper/2007.04612/) (2020)
- [Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs](https://mlsanity.com/paper/2406.16860/) (2024)
- [This Looks Like That: Deep Learning for Interpretable Image Recognition](https://mlsanity.com/paper/1806.10574/) (2018)
- [Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMs](https://mlsanity.com/paper/2401.06209/) (2024)
- [DeepSeek-VL: Towards Real-World Vision-Language Understanding](https://mlsanity.com/paper/2403.05525/) (2024)
- [Network Dissection: Quantifying Interpretability of Deep Visual Representations](https://mlsanity.com/paper/1704.05796/) (2017)
- [Score-CAM: Score-Weighted Visual Explanations for Convolutional Neural Networks](https://mlsanity.com/paper/1910.01279/) (2019)
- [Scaling and evaluating sparse autoencoders](https://mlsanity.com/paper/2406.04093/) (2024)
- [A Benchmark for Interpretability Methods in Deep Neural Networks](https://mlsanity.com/paper/1806.10758/) (2018)
- [Counterfactual Visual Explanations](https://mlsanity.com/paper/1904.07451/) (2019)
- [Post-hoc Concept Bottleneck Models](https://mlsanity.com/paper/2205.15480/) (2022)
- [k-Sparse Autoencoders](https://mlsanity.com/paper/1312.5663/) (2013)
- [Eagle: Exploring The Design Space for Multimodal LLMs with Mixture of Encoders](https://mlsanity.com/paper/2408.15998/) (2024)
- [Natural Language Descriptions of Deep Visual Features](https://mlsanity.com/paper/2201.11114/) (2022)
- [Deformable ProtoPNet: An Interpretable Image Classifier Using Deformable Prototypes](https://mlsanity.com/paper/2111.15000/) (2021)
- [Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factors](https://mlsanity.com/paper/2103.15949/) (2021)
- [Interpreting Deep Visual Representations via Network Dissection](https://mlsanity.com/paper/1711.05611/) (2017)

## Built on by (most cited first, 25 of 25)

- [Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models](https://mlsanity.com/paper/2504.02821/) (2025)
- [Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models](https://mlsanity.com/paper/2603.19233/) (2026)
- [CytoSAE: Interpretable Cell Embeddings for Hematology](https://mlsanity.com/paper/2507.12464/) (2025)
- [Interpretable and Steerable Concept Bottleneck Sparse Autoencoders](https://mlsanity.com/paper/2512.10805/) (2025)
- [Analyzing Hierarchical Structure in Vision Models with Sparse Autoencoders](https://mlsanity.com/paper/2505.15970/) (2025)
- [Causal Interpretation of Sparse Autoencoder Features in Vision](https://mlsanity.com/paper/2509.00749/) (2025)
- [Mind the (Data) Gap: Evaluating Vision Systems in Small Data Applications](https://mlsanity.com/paper/2504.06486/) (2025)
- [Sparse Autoencoders, Again?](https://mlsanity.com/paper/2506.04859/) (2025)
- [Interpreting CFD Surrogates through Sparse Autoencoders](https://mlsanity.com/paper/2507.16069/) (2025)
- [Insights into a radiology-specialised multimodal large language model with sparse autoencoders](https://mlsanity.com/paper/2507.12950/) (2025)
- [Improving Video Diffusion Transformer Training by Multi-Feature Fusion and Alignment from Self-Supervised Vision Encoders](https://mlsanity.com/paper/2509.09547/) (2025)
- [Explaining How Visual, Textual and Multimodal Encoders Share Concepts](https://mlsanity.com/paper/2507.18512/) (2025)
- [GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation Models](https://mlsanity.com/paper/2605.01829/) (2026)
- [Evaluating Vision Language Models (VLMs) for Radiology: A Comprehensive Analysis](https://mlsanity.com/paper/2504.16047/) (2025)
- [Representational Difference Explanations](https://mlsanity.com/paper/2505.23917/) (2025)
- [Pulling Back the Curtain on Deep Networks](https://mlsanity.com/paper/2507.22832/) (2025)
- [Correlates of Image Memorability in Vision Encoders: Activations, Attention Entropy, Patch Uniformity and Autoencoder Losses](https://mlsanity.com/paper/2509.01453/) (2025)
- [Interpreting Video Representations with Spatio-Temporal Sparse Autoencoders](https://mlsanity.com/paper/2604.03919/) (2026)
- [Transcoders Trace Visual Grounding and Hallucinations in Vision-Language Models](https://mlsanity.com/paper/2605.22902/) (2026)
- [A Mechanistic Analysis of Adversarial Fine-tuning of Vision Transformers](https://mlsanity.com/paper/2606.07593/) (2026)
- [Analyzing Visual Aircraft Representations with Sparse Autoencoders](https://mlsanity.com/paper/2606.15468/) (2026)
- [When Confidence Lacks Concepts: Interpretable OOD Detection via Representation Perturbations](https://mlsanity.com/paper/2606.16196/) (2026)
- [EXPOSE: Explainable and Domain-Robust Embeddings from Pathology Vision Foundation Models using Sparse Autoencoders](https://mlsanity.com/paper/2608.28191/) (2026)
- [Verifying the Linear Representation Hypothesis: How Interpretable Are Vision SAEs?](https://mlsanity.com/paper/2609.35020/) (2026)
- [The Independence Prior of SAEs Fragments Visual Concepts](https://mlsanity.com/paper/2610.04112/) (2026)

## Abstract (arXiv)

To truly understand vision models, we must not only interpret their learned features but also validate these interpretations through controlled experiments. While earlier work offers either rich semantics or direct control, few post-hoc tools supply both in a single, model-agnostic procedure. We use sparse autoencoders (SAEs) to bridge this gap; each sparse feature comes with real-image exemplars that reveal its meaning and a decoding vector that can be manipulated to probe its influence on downstream task behavior. By applying our method to widely-used pre-trained vision models, we reveal meaningful differences in the semantic abstractions learned by different pre-training objectives. We then show that a single SAE trained on frozen ViT activations supports patch-level causal edits across tasks (classification and segmentation) all without retraining the ViT or task heads. These qualitative, falsifiable demonstrations position SAEs as a practical bridge between concept discovery and causal probing of vision models. We provide code, demos and models on our project website: https://osu-nlp-group.github.io/saev.