mlsanityData to 7 Oct 2026

SingularClip: Preventing Spectral Collapse to Maintain Plasticity in Continual and Reinforcement Learning

arXiv 2608.18319 ↗First posted 18 Aug 2026cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Neural networks trained on nonstationary tasks frequently lose the ability to fit new targets, a phenomenon referred to as loss of plasticity. We identify a novel source of plasticity loss due to the growing anisotropy of weight matrices' singular values during training, and analyze this phenomenon both empirically and theoretically. To mitigate this issue, we introduce SingularClip, a procedure that periodically clips the singular values of all weight matrices. We show that SingularClip performs strongly against baselines across a range of tasks in both continual supervised learning and deep reinforcement learning.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 52 of its 59 references that are in the map. First posted on arXiv 18 Aug 2026.

Builds on

52 of 59 references.

References · 52

  1. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  2. 2016Layer Normalizationcited by 5,579
  3. 2016Understanding deep learning requires rethinking generalizationcited by 1,657
  4. 2019Root Mean Square Layer Normalizationcited by 1,425
  5. 2018DeepMind Control Suitecited by 935
  6. 2021Deep Reinforcement Learning at the Edge of the Statistical Precipicecited by 781
  7. 2013Exact solutions to the nonlinear dynamics of learning in deep linear neural networkscited by 748
  8. 2013An Empirical Investigation of Catastrophic Forgetting in Gradient-Based Neural Networkscited by 711
  9. 2017The Implicit Bias of Gradient Descent on Separable Datacited by 649
  10. 2016Weight Normalization: A Simple Reparameterization to Accelerate Training of Deep Neural Networkscited by 541
  11. 2017Parseval Networks: Improving Robustness to Adversarial Examplescited by 288
  12. 2018On the Optimization of Deep Networks: Implicit Acceleration by Overparameterizationcited by 280
38 more
  1. 2022The Primacy Bias in Deep Reinforcement Learningcited by 242
  2. 2016Deep Learning without Poor Local Minimacited by 233
  3. 2025LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristicscited by 227
  4. 2018Gradient Descent Happens in a Tiny Subspacecited by 208
  5. 2017Spectral Norm Regularization for Improving the Generalizability of Deep Learningcited by 178
  6. 2023The Dormant Neuron Phenomenon in Deep Reinforcement Learningcited by 171
  7. 2023Understanding plasticity in neural networkscited by 170
  8. 2017Breaking the Softmax Bottleneck: A High-Rank RNN Language Modelcited by 147
  9. 2023Bigger, Better, Faster: Human-level Atari with human-level efficiencycited by 145
  10. 2022Understanding and Preventing Capacity Loss in Reinforcement Learningcited by 124
  11. 2017L2 Regularization versus Batch and Weight Normalizationcited by 121
  12. 2024Bigger, Regularized, Optimistic: scaling for compute and sample-efficient continuous controlcited by 104
  13. 2025The Polar Express: Optimal Matrix Sign Methods and Their Application to the Muon Algorithmcited by 104
  14. 2024Disentangling the Causes of Plasticity Loss in Neural Networkscited by 82
  15. 2023Dichotomy of Early and Late Phase Implicit Biases Can Provably Induce Grokkingcited by 76
  16. 2023Maintaining Plasticity in Continual Learning via Regenerative Regularizationcited by 63
  17. 2024Normalization and effective learning rates in reinforcement learningcited by 55
  18. 2025Hyperspherical Normalization for Scalable Deep Reinforcement Learningcited by 35
  19. 2024Learning Continually by Spectral Regularizationcited by 34
  20. 2021How many degrees of freedom do we need to train deep networks: a loss landscape perspectivecited by 28
  21. 2023Resetting the Optimizer in Deep RL: An Empirical Studycited by 26
  22. 2023Directions of Curvature as an Explanation for Loss of Plasticitycited by 26
  23. 2025Mitigating Plasticity Loss in Continual Reinforcement Learning by Reducing Churncited by 22
  24. 2024Parseval Regularization for Continual Reinforcement Learningcited by 21
  25. 2018Sigsoftmax: Reanalysis of the Softmax Bottleneckcited by 19
  26. 2025Training Transformers with Enforced Lipschitz Constantscited by 18
  27. 2024Dissecting Deep RL with High Update Ratios: Combatting Value Divergencecited by 17
  28. 2025XQC: Well-conditioned Optimization Accelerates Deep Reinforcement Learningcited by 17
  29. 2025Measure gradients, not activations! Enhancing neuronal activity in deep reinforcement learningcited by 14
  30. 2021Correcting Momentum in Temporal Difference Learningcited by 12
  31. 2023Sample-Efficient and Safe Deep Reinforcement Learning via Reset Deep Ensemble Agentscited by 12
  32. 2024Adam on Local Time: Addressing Nonstationarity in RL with Relative Adam Timestepscited by 11
  33. 2025Convergence and Implicit Bias of Gradient Descent on Continual Linear Classificationcited by 8
  34. 2024DASH: Warm-Starting Neural Network Training in Stationary Settings without Loss of Plasticitycited by 7
  35. 2026FIRE: Frobenius-Isometry Reinitialization for Balancing the Stability-Plasticity Tradeoffcited by 7
  36. 2025Spectral Collapse Drives Loss of Plasticity in Deep Continual Learningcited by 5
  37. 2025Barriers for Learning in an Evolving World: Mathematical Understanding of Loss of Plasticitycited by 5
  38. 2026Stable Deep Reinforcement Learning via Isotropic Gaussian Representationscited by 5

50 most cited of 52 All 52 (JSON)

Built on by

Not cited yet.