Interpreting Video Representations with Spatio-Temporal Sparse Autoencoders

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Dokme, Atahan, Vishwanath, Sriram
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866918428336455680
author Dokme, Atahan
Vishwanath, Sriram
author_facet Dokme, Atahan
Vishwanath, Sriram
contents We present the first systematic study of Sparse Autoencoders (SAEs) on video representations. Standard SAEs decompose video into interpretable, monosemantic features but destroy temporal coherence: hard TopK selection produces unstable feature assignments across frames, reducing autocorrelation by 36%. We propose spatio-temporal contrastive objectives and Matryoshka hierarchical grouping that recover and even exceed raw temporal coherence. The contrastive loss weight controls a tunable trade-off between reconstruction and temporal coherence. A systematic ablation on two backbones and two datasets shows that different configurations excel at different goals: reconstruction fidelity, temporal coherence, action discrimination, or interpretability. Contrastive SAE features improve action classification by +3.9% over raw features and text-video retrieval by up to 2.8xR@1. A cross-backbone analysis reveals that standard monosemanticity metrics contain a backbone-alignment artifact: both DINOv2 and VideoMAE produce equally monosemantic features under neutral (CLIP) similarity. Causal ablation confirms that contrastive training concentrates predictive signal into a small number of identifiable features.
format Preprint
id arxiv_https___arxiv_org_abs_2604_03919
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Interpreting Video Representations with Spatio-Temporal Sparse Autoencoders
Dokme, Atahan
Vishwanath, Sriram
Computer Vision and Pattern Recognition
Artificial Intelligence
I.2.10; I.4.7
We present the first systematic study of Sparse Autoencoders (SAEs) on video representations. Standard SAEs decompose video into interpretable, monosemantic features but destroy temporal coherence: hard TopK selection produces unstable feature assignments across frames, reducing autocorrelation by 36%. We propose spatio-temporal contrastive objectives and Matryoshka hierarchical grouping that recover and even exceed raw temporal coherence. The contrastive loss weight controls a tunable trade-off between reconstruction and temporal coherence. A systematic ablation on two backbones and two datasets shows that different configurations excel at different goals: reconstruction fidelity, temporal coherence, action discrimination, or interpretability. Contrastive SAE features improve action classification by +3.9% over raw features and text-video retrieval by up to 2.8xR@1. A cross-backbone analysis reveals that standard monosemanticity metrics contain a backbone-alignment artifact: both DINOv2 and VideoMAE produce equally monosemantic features under neutral (CLIP) similarity. Causal ablation confirms that contrastive training concentrates predictive signal into a small number of identifiable features.
title Interpreting Video Representations with Spatio-Temporal Sparse Autoencoders
topic Computer Vision and Pattern Recognition
Artificial Intelligence
I.2.10; I.4.7
url https://arxiv.org/abs/2604.03919