The reviewed record of science sign in
Pith

arxiv: 2502.12892 · v2 · pith:GIUZVWIM · submitted 2025-02-18 · cs.CV

Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models

Thomas Fel , Ekdeep Singh Lubana , Jacob S. Prince , Matthew Kowal , Victor Boutin , Isabel Papadimitriou , Binxu Wang , Martin Wattenberg
show 2 more authors
Demba Ba Talia Konkle
This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:GIUZVWIMrecord.jsonopen to challenge →

classification cs.CV
keywords dictionariesdictionarysaesarchetypalmodelsconceptconceptsframework
0
0 comments X
read the original abstract

Sparse Autoencoders (SAEs) have emerged as a powerful framework for machine learning interpretability, enabling the unsupervised decomposition of model representations into a dictionary of abstract, human-interpretable concepts. However, we reveal a fundamental limitation: existing SAEs exhibit severe instability, as identical models trained on similar datasets can produce sharply different dictionaries, undermining their reliability as an interpretability tool. To address this issue, we draw inspiration from the Archetypal Analysis framework introduced by Cutler & Breiman (1994) and present Archetypal SAEs (A-SAE), wherein dictionary atoms are constrained to the convex hull of data. This geometric anchoring significantly enhances the stability of inferred dictionaries, and their mildly relaxed variants RA-SAEs further match state-of-the-art reconstruction abilities. To rigorously assess dictionary quality learned by SAEs, we introduce two new benchmarks that test (i) plausibility, if dictionaries recover "true" classification directions and (ii) identifiability, if dictionaries disentangle synthetic concept mixtures. Across all evaluations, RA-SAEs consistently yield more structured representations while uncovering novel, semantically meaningful concepts in large-scale vision models.

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Can neurons speak? Semantic narration of vision at single-cell resolution

    q-bio.NC 2026-06 unverdicted novelty 7.0

    NEURRATOR bridges neural spike trains to frozen CLIP patch embeddings via a learned encoder, then uses a multimodal LM and sparse autoencoder to produce validated natural-language narrations of viewed scenes from Neur...

  2. From Local to Global to Mechanistic: An iERF-Centered Unified Framework for Interpreting Vision Models

    cs.CV 2026-05 unverdicted novelty 7.0

    An iERF-centric framework unifies local, global, and mechanistic interpretability in vision models via SRD for saliency, CAFE for concept anchoring, and ICAT for interlayer attribution.

  3. Can Cross-Layer Transcoders Replace Vision Transformer Activations? An Interpretable Perspective on Vision

    cs.CV 2026-04 unverdicted novelty 7.0

    Cross-Layer Transcoders decompose ViT activations into sparse, depth-aware layer contributions that maintain zero-shot accuracy and enable faithful attribution of the final representation.

  4. Mechanistic Interpretability with Sparse Autoencoder Neural Operators

    cs.LG 2025-09 unverdicted novelty 7.0

    SAE-NOs extend sparse autoencoders to function spaces via Fourier neural operators with concept and domain sparsity, learning localized patterns more efficiently and generalizing across discretizations on vision data.

  5. Beyond Interpretability: When, Why, and How Sparse Autoencoders Enable Label-Free Visual Steering

    cs.CV 2025-06 unverdicted novelty 6.0

    VS2 constructs steering vectors from sparse SAE features on unlabeled in-domain activations to improve zero-shot accuracy of CLIP models by 0.93-4.12% on CIFAR-100, CUB-200, and Tiny-ImageNet while remaining forward-p...