mlsanityData to 14 Oct 2026

LayerLock: Non-collapsing Representation Learning with Progressive Freezing

arXiv 2509.10156 ↗First posted 12 Sep 2025cs.CVcited by 4

Not on the map. Why?

Abstract (arXiv)

We introduce LayerLock, a simple yet effective approach for self-supervised visual representation learning, that gradually transitions from pixel to latent prediction through progressive layer freezing. First, we make the observation that during training of video masked-autoencoding (MAE) models, ViT layers converge in the order of their depth: shallower layers converge early, deeper layers converge late. We then show that this observation can be exploited to accelerate standard MAE by progressively freezing the model according to an explicit schedule, throughout training. Furthermore, this same schedule can be used in a simple and scalable approach to latent prediction that does not suffer from "representation collapse". We apply our proposed approach, LayerLock, to large models of up to 4B parameters with results surpassing those of non-latent masked prediction on the 4DS perception suite.

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Builds on 41 of its 56 references that are in the map. First posted on arXiv 12 Sep 2025.

Builds on

41 of 56 references.

References · 41

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  2. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
  3. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  4. 2018Representation Learning with Contrastive Predictive Codingcited by 7,777
  5. 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,513
  6. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,403
  7. 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,968
  8. 2020Bootstrap your own latent: A new approach to self-supervised Learningcited by 4,750
  9. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,703
  10. 2017ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenescited by 3,117
  11. 2020Unsupervised Learning of Visual Features by Contrasting Cluster Assignmentscited by 2,857
  12. 2021Barlow Twins: Self-Supervised Learning via Redundancy Reductioncited by 1,838
29 more
  1. 2021Vision Transformers for Dense Predictioncited by 1,803
  2. 2024Depth Anything: Unleashing the Power of Large-Scale Unlabeled Datacited by 1,360
  3. 2022VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Trainingcited by 1,322
  4. 2017The "something something" video database for learning and evaluating visual common sensecited by 1,103
  5. 2018Deep Clustering for Unsupervised Learning of Visual Featurescited by 1,028
  6. 2021iBOT: Image BERT Pre-Training with Online Tokenizercited by 903
  7. 2023Scaling Vision Transformers to 22 Billion Parameterscited by 694
  8. 2024Revisiting Feature Prediction for Learning Visual Representations from Videocited by 531
  9. 2019Self-labelling via simultaneous clustering and representation learningcited by 528
  10. 2023VideoMAE V2: Scaling Video Masked Autoencoders with Dual Maskingcited by 528
  11. 2017Time-Contrastive Networks: Self-Supervised Learning from Videocited by 459
  12. 2020Spatiotemporal Contrastive Video Representation Learningcited by 372
  13. 2021TCLR: Temporal Contrastive Learning for Video Representationcited by 136
  14. 2022VICRegL: Self-Supervised Learning of Local Visual Featurescited by 118
  15. 2024Scalable Pre-training of Large Autoregressive Image Modelscited by 115
  16. 2021Divide and Contrast: Self-supervised Learning from Uncurated Datacited by 85
  17. 2022BYOL-Explore: Exploration by Bootstrapped Predictioncited by 73
  18. 2024Learning by Reconstruction Produces Uninformative Features For Perceptioncited by 55
  19. 2017FreezeOut: Accelerate Training by Progressively Freezing Layerscited by 54
  20. 2022Understanding Collapse in Non-Contrastive Siamese Representation Learningcited by 42
  21. 2023Is ImageNet worth 1 video? Learning strong image encoders from 1 long unlabelled videocited by 40
  22. 2022On the Importance of Asymmetry for Siamese Representation Learningcited by 39
  23. 2023Time Does Tell: Self-Supervised Time-Tuning of Dense Image Representationscited by 35
  24. 2015Spatio-temporal video autoencoder with differentiable memorycited by 32
  25. 2024Scaling 4D Representationscited by 30
  26. 2022Self-supervised video pretraining yields robust and more human-aligned visual representationscited by 20
  27. 2022Motion Sensitive Contrastive Learning for Self-supervised Video Representationcited by 11
  28. 2024Moving Off-the-Grid: Scene-Grounded Video Representationscited by 11
  29. 2024Many Perception Tasks are Highly Redundant Functions of their Input Datacited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026Progressive Residual Warmup for Language Model Pretrainingcited by 1
  2. 2026Staged Depth Training: A Representation Curriculum for PINNscited by 0
  3. 2026When Can Attention Heads Be Statically Defined?cited by 0
  4. 2026Growing to See: Efficient Pretraining of Self-Supervised Vision Transformerscited by 0