# Understanding Video Transformers via Universal Concept Discovery

arXiv: https://arxiv.org/abs/2401.10831 · HTML: https://mlsanity.com/paper/2401.10831/

## Key facts

- Cited by 23 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 51 of its 78 references that are in the map.
- First posted on arXiv 19 Jan 2024.

## 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 51)

- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Segment Anything](https://mlsanity.com/paper/2304.02643/) (2023)
- [Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset](https://mlsanity.com/paper/1705.07750/) (2017)
- [Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets](https://mlsanity.com/paper/2311.15127/) (2023)
- [LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs](https://mlsanity.com/paper/2111.02114/) (2021)
- [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://mlsanity.com/paper/2203.12602/) (2022)
- [Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)](https://mlsanity.com/paper/1711.11279/) (2017)
- [The "something something" video database for learning and evaluating visual common sense](https://mlsanity.com/paper/1706.04261/) (2017)
- [Toy Models of Superposition](https://mlsanity.com/paper/2209.10652/) (2022)
- [In-context Learning and Induction Heads](https://mlsanity.com/paper/2209.11895/) (2022)
- [Object-Centric Learning with Slot Attention](https://mlsanity.com/paper/2006.15055/) (2020)
- [Vision Transformers Need Registers](https://mlsanity.com/paper/2309.16588/) (2023)
- [Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Pruned](https://mlsanity.com/paper/1905.09418/) (2019)
- [RISE: Randomized Input Sampling for Explanation of Black-box Models](https://mlsanity.com/paper/1806.07421/) (2018)
- [Are Sixteen Heads Really Better than One?](https://mlsanity.com/paper/1905.10650/) (2019)
- [Network Dissection: Quantifying Interpretability of Deep Visual Representations](https://mlsanity.com/paper/1704.05796/) (2017)
- [Transformer Interpretability Beyond Attention Visualization](https://mlsanity.com/paper/2012.09838/) (2020)
- [VideoMAE V2: Scaling Video Masked Autoencoders with Dual Masking](https://mlsanity.com/paper/2303.16727/) (2023)
- [InternVideo: General Video Foundation Models via Generative and Discriminative Learning](https://mlsanity.com/paper/2212.03191/) (2022)
- [Masked Autoencoders As Spatiotemporal Learners](https://mlsanity.com/paper/2205.09113/) (2022)
- [Kubric: A scalable dataset generator](https://mlsanity.com/paper/2203.03570/) (2022)
- [Deep ViT Features as Dense Visual Descriptors](https://mlsanity.com/paper/2112.05814/) (2021)
- [Generic Attention-model Explainability for Interpreting Bi-Modal and Encoder-Decoder Transformers](https://mlsanity.com/paper/2103.15679/) (2021)
- [On the Relationship between Self-Attention and Convolutional Layers](https://mlsanity.com/paper/1911.03584/) (2019)
- [On Completeness-aware Concept-Based Explanations in Deep Neural Networks](https://mlsanity.com/paper/1910.07969/) (2019)
- [Human Action Recognition and Prediction: A Survey](https://mlsanity.com/paper/1806.11230/) (2018)
- [Bridging the Gap to Real-World Object-Centric Learning](https://mlsanity.com/paper/2209.14860/) (2022)
- [CRAFT: Concept Recursive Activation FacTorization for Explainability](https://mlsanity.com/paper/2211.10154/) (2022)
- [REPAIR: Removing Representation Bias by Dataset Resampling](https://mlsanity.com/paper/1904.07911/) (2019)
- [Understanding The Robustness in Vision Transformers](https://mlsanity.com/paper/2204.12451/) (2022)

## Built on by (most cited first, 23 of 23)

- [Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models](https://mlsanity.com/paper/2502.12892/) (2025)
- [Universal Sparse Autoencoders: Interpretable Cross-Model Concept Alignment](https://mlsanity.com/paper/2502.03714/) (2025)
- [From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit](https://mlsanity.com/paper/2506.03093/) (2025)
- [Visual Concept Connectome (VCC): Open World Concept Discovery and their Interlayer Connections in Deep Models](https://mlsanity.com/paper/2404.02233/) (2024)
- [Interpreting Physics in Video World Models](https://mlsanity.com/paper/2602.07050/) (2026)
- [Prisma: An Open Source Toolkit for Mechanistic Interpretability in Vision and Video](https://mlsanity.com/paper/2504.19475/) (2025)
- [Concept Influence: Leveraging Interpretability to Improve Performance and Efficiency in Training Data Attribution](https://mlsanity.com/paper/2602.14869/) (2026)
- [Representational Similarity via Interpretable Visual Concepts](https://mlsanity.com/paper/2503.15699/) (2025)
- [Bi-Orthogonal Factor Decomposition for Vision Transformers](https://mlsanity.com/paper/2601.05328/) (2026)
- [Structuring Sparsity: Block-Sparse Featurizers Capture Visual Concept Manifolds](https://mlsanity.com/paper/2606.25234/) (2026)
- [Looking Beyond the Obvious: A Survey on Abstract Concept Recognition for Video Understanding](https://mlsanity.com/paper/2508.20765/) (2025)
- [A Geometric Unification of Concept Learning with Concept Cones](https://mlsanity.com/paper/2512.07355/) (2025)
- [Disentangled Concepts Speak Louder Than Words: Explainable Video Action Recognition](https://mlsanity.com/paper/2511.03725/) (2025)
- [Characterizing Universal Object Representations Across Vision Models](https://mlsanity.com/paper/2605.13675/) (2026)
- [PCBEAR: Pose Concept Bottleneck for Explainable Action Recognition](https://mlsanity.com/paper/2504.13140/) (2025)
- [Concepts in Motion: Temporal Concept Bottleneck Model for Interpretable Video Classification](https://mlsanity.com/paper/2509.20899/) (2025)
- [Neurosymbolic Framework for Concept-Driven Logical Reasoning in Skeleton-Based Human Action Recognition](https://mlsanity.com/paper/2605.07140/) (2026)
- [Relevance-driven Input Dropout: an Explanation-guided Regularization Technique](https://mlsanity.com/paper/2505.21595/) (2025)
- [Learning reusable concepts across different egocentric video understanding tasks](https://mlsanity.com/paper/2505.24690/) (2025)
- [ConceptFlow: Hierarchical and Fine-grained Concept-Based Explanation for Convolutional Neural Networks](https://mlsanity.com/paper/2509.18147/) (2025)
- [Bias Leaves a Gradient Trail: Label-Free Bias Identification via Gradient Probes on Concept Decompositions](https://mlsanity.com/paper/2605.28780/) (2026)
- [What, Where, and How: Probing Spatiotemporal Representations in Video Foundation Models](https://mlsanity.com/paper/2609.01551/) (2026)
- [Investigating Spatiotemporal Redundancy in Video Transformer for Collision Anticipation](https://mlsanity.com/paper/2610.04727/) (2026)

## Abstract (arXiv)

This paper studies the problem of concept-based interpretability of transformer representations for videos. Concretely, we seek to explain the decision-making process of video transformers based on high-level, spatiotemporal concepts that are automatically discovered. Prior research on concept-based interpretability has concentrated solely on image-level tasks. Comparatively, video models deal with the added temporal dimension, increasing complexity and posing challenges in identifying dynamic concepts over time. In this work, we systematically address these challenges by introducing the first Video Transformer Concept Discovery (VTCD) algorithm. To this end, we propose an efficient approach for unsupervised identification of units of video transformer representations - concepts, and ranking their importance to the output of a model. The resulting concepts are highly interpretable, revealing spatio-temporal reasoning mechanisms and object-centric representations in unstructured video models. Performing this analysis jointly over a diverse set of supervised and self-supervised representations, we discover that some of these mechanism are universal in video transformers. Finally, we show that VTCD can be used for fine-grained action recognition and video object segmentation.