# Rethinking JEPA: Compute-Efficient Video SSL with Frozen Teachers

arXiv: https://arxiv.org/abs/2509.24317 · HTML: https://mlsanity.com/paper/2509.24317/

## Key facts

- Cited by 12 papers in the map (data through 14 Oct 2026).
- Builds on 49 of its 56 references that are in the map.
- First posted on arXiv 29 Sep 2025.

## Builds on (most cited first, 30 of 49)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [ImageNet Large Scale Visual Recognition Challenge](https://mlsanity.com/paper/1409.0575/) (2014)
- [Segment Anything](https://mlsanity.com/paper/2304.02643/) (2023)
- [DINOv2: Learning Robust Visual Features without Supervision](https://mlsanity.com/paper/2304.07193/) (2023)
- [Masked Autoencoders Are Scalable Vision Learners](https://mlsanity.com/paper/2111.06377/) (2021)
- [Emerging Properties in Self-Supervised Vision Transformers](https://mlsanity.com/paper/2104.14294/) (2021)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [Bootstrap your own latent: A new approach to self-supervised Learning](https://mlsanity.com/paper/2006.07733/) (2020)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [Training Compute-Optimal Large Language Models](https://mlsanity.com/paper/2203.15556/) (2022)
- [The Kinetics Human Action Video Dataset](https://mlsanity.com/paper/1705.06950/) (2017)
- [Barlow Twins: Self-Supervised Learning via Redundancy Reduction](https://mlsanity.com/paper/2103.03230/) (2021)
- [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://mlsanity.com/paper/2203.12602/) (2022)
- [Self-training with Noisy Student improves ImageNet classification](https://mlsanity.com/paper/1911.04252/) (2019)
- [SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features](https://mlsanity.com/paper/2502.14786/) (2025)
- [The "something something" video database for learning and evaluating visual common sense](https://mlsanity.com/paper/1706.04261/) (2017)
- [VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning](https://mlsanity.com/paper/2105.04906/) (2021)
- [Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture](https://mlsanity.com/paper/2301.08243/) (2023)
- [HowTo100M: Learning a Text-Video Embedding by Watching Hundred Million Narrated Video Clips](https://mlsanity.com/paper/1906.03327/) (2019)
- [Cosmos World Foundation Model Platform for Physical AI](https://mlsanity.com/paper/2501.03575/) (2025)
- [V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning](https://mlsanity.com/paper/2506.09985/) (2025)
- [PaLI: A Jointly-Scaled Multilingual Language-Image Model](https://mlsanity.com/paper/2209.06794/) (2022)
- [Revisiting Feature Prediction for Learning Visual Representations from Video](https://mlsanity.com/paper/2404.08471/) (2024)
- [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)
- [Panda-70M: Captioning 70M Videos with Multiple Cross-Modality Teachers](https://mlsanity.com/paper/2402.19479/) (2024)
- [DINO-WM: World Models on Pre-trained Visual Features enable Zero-shot Planning](https://mlsanity.com/paper/2411.04983/) (2024)
- [Perception Encoder: The best visual embeddings are not at the output of the network](https://mlsanity.com/paper/2504.13181/) (2025)

## Built on by (most cited first, 12 of 12)

- [World Action Models: The Next Frontier in Embodied AI](https://mlsanity.com/paper/2605.12090/) (2026)
- [Causal-JEPA: Learning World Models through Object-Level Latent Masking](https://mlsanity.com/paper/2602.11389/) (2026)
- [Seeing Beyond Words: Self-Supervised Visual Learning for Multimodal Large Language Models](https://mlsanity.com/paper/2512.15885/) (2025)
- [AeroJEPA: Learning Semantic Latent Representations for Scalable 3D Aerodynamic Field Modeling](https://mlsanity.com/paper/2605.05586/) (2026)
- [US-JEPA: A Joint Embedding Predictive Architecture for Ultrasound](https://mlsanity.com/paper/2602.19322/) (2026)
- [Latent Clarity: Bridging World-Model Kinematics to Semantic Manifolds for Video Anomaly Anticipation](https://mlsanity.com/paper/2607.03558/) (2026)
- [Scalable and Efficient Joint Spiking Embedding Predictive Architecture for Large-Scale Dynamic Graphs](https://mlsanity.com/paper/2607.18412/) (2026)
- [GOT-JEPA: Generic Object Tracking with Model Adaptation and Occlusion Handling using Joint-Embedding Predictive Architecture](https://mlsanity.com/paper/2602.14771/) (2026)
- [Mask What Matters: Saliency-Guided Video Self-Supervised Learning for Autonomous Driving](https://mlsanity.com/paper/2608.17178/) (2026)
- [Mask-Aware Execution for Efficient JEPA Training](https://mlsanity.com/paper/2609.22674/) (2026)
- [JEPA Learns What the Mask Leaves Unrecoverable](https://mlsanity.com/paper/2609.32481/) (2026)
- [TT-VidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining](https://mlsanity.com/paper/2609.33419/) (2026)

## Abstract (arXiv)

Video Joint Embedding Predictive Architectures (V-JEPA) learn generalizable off-the-shelf video representation by predicting masked regions in latent space with an exponential moving average (EMA)-updated teacher. While EMA prevents representation collapse, it complicates scalable model selection and couples teacher and student architectures. We revisit masked-latent prediction and show that a frozen teacher suffices. Concretely, we (i) train a target encoder with a simple pixel-reconstruction objective under V-JEPA masking, then (ii) freeze it and train a student to predict the teacher's latents on masked regions. This leads to a two-stage, unregularized scheme that we refer to as SALT (Static-teacher Asymmetric Latent Training). SALT decouples optimization into pixel reconstruction (teacher) and masked latent prediction (student), increasing transparency, efficiency, and scalability while preserving the ability of representation to generalize under frozen evaluation. Empirically, our student models outperform recently proposed V-JEPA 2 encoders under frozen backbone evaluation across diverse benchmarks. They are also more compute-optimal: at matched pretraining FLOPs, our method achieves higher probing accuracy, and its scaling curves dominate V-JEPA's accuracy-FLOPs Pareto frontier. Finally, we find that student quality is remarkably robust to teacher quality: high-performing students emerge even with small, sub-optimal teachers. This points to a compute budget allocation that should overwhelmingly favor the student. These results position SALT as a simple, scalable, and compute-efficient alternative to EMA-based self-distillation for video representation learning.