mlsanityData to 14 Oct 2026

LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics

arXiv 2511.08544 ↗First posted 11 Nov 2025cs.LG cs.AI cs.CVcited by 254

Not on the map. Why?

Cited by 102 in World models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Learning manipulable representations of the world and its dynamics is central to AI. Joint-Embedding Predictive Architectures (JEPAs) offer a promising blueprint, but lack of practical guidance and theory has led to ad-hoc R&D. We present a comprehensive theory of JEPAs and instantiate it in bf LeJEPA, a lean, scalable, and theoretically grounded training objective. First, we identify the isotropic Gaussian as the optimal distribution that JEPAs' embeddings should follow to minimize downstream prediction risk. Second, we introduce a novel objective--bf Sketched Isotropic Gaussian Regularization (SIGReg)--to constrain embeddings to reach that ideal distribution. Combining the JEPA predictive loss with SIGReg yields LeJEPA with numerous theoretical and practical benefits: (i) single trade-off hyperparameter, (ii) linear time and memory complexity, (iii) stability across hyper-parameters, architectures (ResNets, ViTs, ConvNets) and domains, (iv) heuristics-free, e.g., no stop-gradient, no teacher-student, no hyper-parameter schedulers, and (v) distributed training-friendly implementation requiring only ≈50 lines of code. Our empirical validation covers 10+ datasets, 60+ architectures, all with varying scales and domains. As an example, using imagenet-1k for pretraining and linear evaluation with frozen backbone, LeJEPA reaches 79% with a ViT-H/14. We hope that the simplicity and theory-friendly ecosystem offered by LeJEPA will reestablish self-supervised pre-training as a core pillar of AI research (hrefhttps://github.com/rbalestr-lab/lejepaGitHub repo).

Key facts. Cited by 254 papers in the map (data through 14 Oct 2026). Builds on 53 of its 121 references that are in the map. First posted on arXiv 11 Nov 2025.

Builds on

53 of 121 references.

References · 53

  1. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
  2. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  3. 2018Representation Learning with Contrastive Predictive Codingcited by 7,777
  4. 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,513
  5. 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,968
  6. 2020Bootstrap your own latent: A new approach to self-supervised Learningcited by 4,750
  7. 2021An Empirical Study of Training Self-Supervised Vision Transformerscited by 1,616
  8. 2025DINOv3cited by 1,414
  9. 2020Big Self-Supervised Models are Strong Semi-Supervised Learnerscited by 1,380
  10. 2018Learning deep representations by mutual information estimation and maximizationcited by 1,273
  11. 2018Averaging Weights Leads to Wider Optima and Better Generalizationcited by 1,209
  12. 2021VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learningcited by 1,057
38 more
  1. 2023Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecturecited by 1,039
  2. 2023Vision Transformers Need Registerscited by 871
  3. 2024DROID: A Large-Scale In-The-Wild Robot Manipulation Datasetcited by 852
  4. 2020What Makes for Good Views for Contrastive Learning?cited by 846
  5. 2020Prevalence of Neural Collapse during the terminal phase of deep learning trainingcited by 719
  6. 2020Self-supervised Learning: Generative or Contrastivecited by 694
  7. 2019On Variational Bounds of Mutual Informationcited by 559
  8. 2014Semi-Supervised Learning with Deep Generative Modelscited by 496
  9. 2019Sliced Score Matching: A Scalable Approach to Density and Score Estimationcited by 384
  10. 2021Understanding Dimensional Collapse in Contrastive Self-supervised Learningcited by 378
  11. 2020Whitening for Self-Supervised Representation Learningcited by 279
  12. 2021Understanding self-supervised Learning Dynamics without Contrastive Pairscited by 256
  13. 2023A Cookbook of Self-Supervised Learningcited by 250
  14. 2019Scaling and Benchmarking Self-Supervised Visual Representation Learningcited by 201
  15. 2018Formal Limitations on the Measurement of Mutual Informationcited by 189
  16. 2022RankMe: Assessing the downstream performance of pretrained self-supervised representations by their rankcited by 153
  17. 2022Contrastive and Non-Contrastive Self-Supervised Learning Recover Global and Local Spectral Embedding Methodscited by 149
  18. 2016A Kernel Test of Goodness of Fitcited by 143
  19. 2019Differentiable Ranks and Sorting using Optimal Transportcited by 113
  20. 2019Stochastic Optimization of Sorting Networks via Continuous Relaxationscited by 111
  21. 2018Noise Contrastive Estimation and Negative Sampling for Conditional Models: Consistency and Statistical Efficiencycited by 105
  22. 2023To Compress or Not to Compress- Self-Supervised Learning and Information Theory: A Reviewcited by 72
  23. 2025Scaling Language-Free Visual Representation Learningcited by 66
  24. 2022The Hidden Uniform Cluster Prior in Self-Supervised Learningcited by 58
  25. 2024Learning by Reconstruction Produces Uninformative Features For Perceptioncited by 55
  26. 2024Automatic Data Curation for Self-Supervised Learning: A Clustering-Based Approachcited by 50
  27. 2022On the Importance of Asymmetry for Siamese Representation Learningcited by 39
  28. 2023Energy-Based Sliced Wasserstein Distancecited by 39
  29. 2025Joint Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self Supervised Learningcited by 36
  30. 2023LiDAR: Sensing Linear Probing Performance in Joint Embedding SSL Architecturescited by 27
  31. 2017Distance and Similarity Measures Effect on the Performance of K-Nearest Neighbor Classifier -- A Reviewcited by 26
  32. 2022Monotonic Differentiable Sorting Networkscited by 25
  33. 2023Matrix Information Theory for Self-Supervised Learningcited by 21
  34. 2022Toward a Geometrical Understanding of Self-supervised Contrastive Learningcited by 17
  35. 2022What Do We Maximize in Self-Supervised Learning?cited by 15
  36. 2025Gaussian Embeddings: How JEPAs Secretly Learn Your Data Densitycited by 13
  37. 2023Stochastic positional embeddings improve masked image modelingcited by 12
  38. 2019Computing Extremely Accurate Quantiles Using t-Digestscited by 10

50 most cited of 53 All 53 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 6; 2026: 248.

254 papers.

Sort:

By area:

  1. 2026LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixelscited by 273
  2. 2026V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learningcited by 117
  3. 2025VL-JEPA: Joint Embedding Predictive Architecture for Vision-languagecited by 68
  4. 2026Temporal Straightening for Latent Planningcited by 38
  5. 2026When Does LeJEPA Learn a World Model?cited by 30
  6. 2026stable-worldmodel: A Platform for Reproducible World Modeling Research and Evaluationcited by 20
  7. 2026Sensorimotor World Models: Perception for Action via Inverse Dynamicscited by 20
  8. 2026AdaJEPA: An Adaptive Latent World Modelcited by 20
  9. 2026Predictive but Not Plannable: RC-aux for Latent World Modelscited by 19
  10. 2026InfoNCE Induces Gaussian Distributioncited by 16
  11. 2026VISReg: Variance-Invariance-Sketching Regularization for JEPA trainingcited by 16
  12. 2026Latent Geometry Beyond Search: Amortizing Planning in World Modelscited by 15
38 more
  1. 2026Fast LeWorldModelcited by 15
  2. 2026Delta-JEPA: Learning Action-Sensitive World Models via Latent Difference Decodingcited by 15
  3. 2026A Lightweight Library for Energy-Based Joint-Embedding Predictive Architecturescited by 14
  4. 2026Learning Invariant Visual Representations for Planning with Joint-Embedding Predictive World Modelscited by 13
  5. 2025Closing the Train-Test Gap in World Models for Gradient-Based Planningcited by 11
  6. 2026Subspace-Decomposed JEPAs: Disentangling Progression and Content in Latent World Modelscited by 11
  7. 2026JEPA-DNA: Grounding Genomic Foundation Models through Joint-Embedding Predictive Architecturescited by 9
  8. 2026SkyJEPA: Learning Long-Horizon World Models for Zero-Shot Sim-to-Real Control of Quadrotorscited by 9
  9. 2026A Control Theory of Predictability in Latent World Modelscited by 9
  10. 2026On the Identifiability of Controlled World Modelscited by 9
  11. 2025KerJEPA: Kernel Discrepancies for Euclidean Self-Supervised Learningcited by 8
  12. 2026Video-Mirai: Autoregressive Video Diffusion Models Need Foresightcited by 8
  13. 2026Temporal-Distance JEPA: Plan-Aware Representation Learning for Latent World Model Predictive Controlcited by 8
  14. 2026Is Forward Prediction Enough? Physical State Grounding for JEPA World Modelscited by 8
  15. 2026LpWM: A Case for Sparse Representations in World Modelscited by 8
  16. 2026Decision-Metric Alignment in Latent World Models: Diagnostics and Action-Conditioned Objectives for MPC Planningcited by 7
  17. 2026Curia-2: Scaling Self-Supervised Learning for Radiology Foundation Modelscited by 6
  18. 2026Orca: The World is in Your Mindcited by 6
  19. 2026PhyLatent: Learning Dynamics-Relevant Representations for JEPA World Modelscited by 6
  20. 2026Self-Supervised JEPA-based World Models for LiDAR Occupancy Completion and Forecastingcited by 5
  21. 2026Stable Deep Reinforcement Learning via Isotropic Gaussian Representationscited by 5
  22. 2026Nano World Models: A Minimalist Implementation of Future Video Predictioncited by 5
  23. 2026SPHERE-JEPA: Spherical Prediction with Homogeneous Embeddingscited by 5
  24. 2026You Don't Need Strong Assumptions: Visual Representation Learning via Temporal Differencescited by 5
  25. 2026A Generalization Theory for JEPA-Based World Modelscited by 5
  26. 2026LeNEPA: No-Augmentation Next-Latent Prediction for Time-Series Representation Learningcited by 5
  27. 2026ACID: Action Consistency via Inverse Dynamics for Planning with World Modelscited by 5
  28. 2026QQWorld: Quantile-Quantile Matching for World Model Regularizationcited by 5
  29. 2026No Gaussian Required: Contrastive Inverse Dynamics for JEPA World Modelscited by 5
  30. 2025DSeq-JEPA: Discriminative Sequential Joint-Embedding Predictive Architecturecited by 4
  31. 2026Laya: A LeJEPA Approach to EEG via Latent Prediction over Reconstructioncited by 4
  32. 2026LuMamba: Latent Unified Mamba for Electrode Topology-Invariant and Efficient EEG Modelingcited by 4
  33. 2026Var-JEPA: A Variational Formulation of the Joint-Embedding Predictive Architecture - Bridging Predictive and Generative Self-Supervised Learningcited by 4
  34. 2026Policy-Guided World Model Planning for Language-Conditioned Visual Navigationcited by 4
  35. 2026Understanding Self-Supervised Learning via Latent Distribution Matchingcited by 4
  36. 2026AeroJEPA: Learning Semantic Latent Representations for Scalable 3D Aerodynamic Field Modelingcited by 4
  37. 2026Crys-JEPA: Accelerating Crystal Discovery via Embedding Screening and Generative Refinementcited by 4
  38. 2026Imperfect World Models are Exploitablecited by 4

50 most cited of 254 All 254 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024How JEPA Avoids Noisy Features: The Implicit Bias of Deep Linear Self Distillation Networks
  2. 2026VJEPA: Variational Joint Embedding Predictive Architectures as Probabilistic World Models
  3. 2026UniJEPA: A Unified Joint-Embedding Predictive Architecture for Task-Agnostic Visual World Modeling
  4. 2025JEPA for RL: Investigating Joint-Embedding Predictive Architectures for Reinforcement Learning
  5. 2026Sub-JEPA: Subspace Gaussian Regularization for Stable End-to-End World Models
  6. 2022Joint Embedding Predictive Architectures Focus on Slow Features
  7. 2025seq-JEPA: Autoregressive Predictive Learning of Invariant-Equivariant World Models
  8. 2026Factorized Latent Dynamics for Video JEPA: An Empirical Study of Auxiliary Objectives