mlsanityData to 7 Oct 2026

Maintaining Plasticity in Deep Continual Learning

arXiv 2306.13812 ↗First posted 23 Jun 2023cs.LGcited by 42

Not on the map. Why?

Abstract (arXiv)

Modern deep-learning systems are specialized to problem settings in which training occurs once and then never again, as opposed to continual-learning settings in which training occurs continually. If deep-learning systems are applied in a continual learning setting, then it is well known that they may fail to remember earlier examples. More fundamental, but less well known, is that they may also lose their ability to learn on new examples, a phenomenon called loss of plasticity. We provide direct demonstrations of loss of plasticity using the MNIST and ImageNet datasets repurposed for continual learning as sequences of tasks. In ImageNet, binary classification performance dropped from 89% accuracy on an early task down to 77%, about the level of a linear network, on the 2000th task. Loss of plasticity occurred with a wide range of deep network architectures, optimizers, activation functions, batch normalization, dropout, but was substantially eased by L2-regularization, particularly when combined with weight perturbation. Further, we introduce a new algorithm -- continual backpropagation -- which slightly modifies conventional backpropagation to reinitialize a small fraction of less-used units after each example and appears to maintain plasticity indefinitely.

Key facts. Cited by 42 papers in the map (data through 7 Oct 2026). Builds on 53 of its 98 references that are in the map. First posted on arXiv 23 Jun 2023.

Builds on

53 of 98 references.

References · 53

  1. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  2. 2020Language Models are Few-Shot Learnerscited by 32,199
  3. 2017Proximal Policy Optimization Algorithmscited by 14,441
  4. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  5. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  6. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,861
  7. 2017Squeeze-and-Excitation Networkscited by 4,763
  8. 2016Overcoming catastrophic forgetting in neural networkscited by 4,718
  9. 2021Zero-Shot Text-to-Image Generationcited by 4,276
  10. 2015Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classificationcited by 3,590
  11. 2015Continuous control with deep reinforcement learningcited by 3,279
  12. 2015Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Codingcited by 2,291
38 more
  1. 2018The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networkscited by 2,250
  2. 2016Asynchronous Methods for Deep Reinforcement Learningcited by 2,218
  3. 2016Progressive Neural Networkscited by 1,332
  4. 2018Continual Lifelong Learning with Neural Networks: A Reviewcited by 1,313
  5. 2015Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)cited by 1,061
  6. 2018Riemannian Walk for Incremental Learning: Understanding Forgetting and Intransigencecited by 734
  7. 2013An Empirical Investigation of Catastrophic Forgetting in Gradient-Based Neural Networkscited by 711
  8. 2017Lifelong Learning with Dynamically Expandable Networkscited by 689
  9. 2013Speech Recognition with Deep Recurrent Neural Networkscited by 666
  10. 2018Learning to Learn without Forgetting by Maximizing Transfer and Minimizing Interferencecited by 511
  11. 2019Online Continual Learning with Maximally Interfered Retrievalcited by 376
  12. 2017A Downsampled Variant of ImageNet as an Alternative to the CIFAR datasetscited by 374
  13. 2018How Does Batch Normalization Help Optimization?cited by 355
  14. 2017Scalable Training of Artificial Neural Networks with Adaptive Sparse Connectivity inspired by Network Sciencecited by 325
  15. 2015LSTM: A Search Space Odysseycited by 315
  16. 2018Learning to Adapt in Dynamic, Real-World Environments Through Meta-Reinforcement Learningcited by 270
  17. 2022The Primacy Bias in Deep Reinforcement Learningcited by 242
  18. 2016Network Trimming: A Data-Driven Neuron Pruning Approach towards Efficient Deep Architecturescited by 220
  19. 2019Meta-Learning Representations for Continual Learningcited by 189
  20. 2021On the Origin of Implicit Regularization in Stochastic Gradient Descentcited by 183
  21. 2021Chasing Sparsity in Vision Transformers: An End-to-End Explorationcited by 181
  22. 2023The Dormant Neuron Phenomenon in Deep Reinforcement Learningcited by 171
  23. 2023Understanding plasticity in neural networkscited by 170
  24. 2018Understanding Batch Normalizationcited by 157
  25. 2023Bigger, Better, Faster: Human-level Atari with human-level efficiencycited by 145
  26. 2015Adding Gradient Noise Improves Learning for Very Deep Networkscited by 144
  27. 2017Deep Rewiring: Training very sparse deep networkscited by 136
  28. 2020Implicit Under-Parameterization Inhibits Data-Efficient Deep Reinforcement Learningcited by 134
  29. 2022Understanding and Preventing Capacity Loss in Reinforcement Learningcited by 124
  30. 2020Implicit Regularization in Deep Learning May Not Be Explainable by Normscited by 118
  31. 2023Loss of Plasticity in Continual Deep Reinforcement Learningcited by 118
  32. 2019Continual Learning via Neural Pruningcited by 101
  33. 2020Efficient Continual Learning with Modular Networks and Task-Driven Priorscited by 87
  34. 2020Transient Non-Stationarity and Generalisation in Deep Reinforcement Learningcited by 75
  35. 2023Deep Reinforcement Learning with Plasticity Injectioncited by 71
  36. 2020Dynamic Sparse Training: Find Efficient Sparse Network From Scratch With Trainable Masked Layerscited by 66
  37. 2021Dynamic Sparse Training for Deep Reinforcement Learningcited by 49
  38. 2021A study on the plasticity of neural networkscited by 44

50 most cited of 53 All 53 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 5; 2024: 13; 2025: 11; 2026: 13.

42 papers.

Sort:

By area:

  1. 2023Maintaining Plasticity in Continual Learning via Regenerative Regularizationcited by 63
  2. 2024Addressing Loss of Plasticity and Catastrophic Forgetting in Continual Learningcited by 49
  3. 2025Hyperspherical Normalization for Scalable Deep Reinforcement Learningcited by 35
  4. 2024Learning Continually by Spectral Regularizationcited by 34
  5. 2023Revisiting Plasticity in Visual Reinforcement Learning: Data, Modules and Training Stagescited by 27
  6. 2024No Representation, No Trust: Connecting Representation, Collapse, and Trust Issues in PPOcited by 27
  7. 2023Directions of Curvature as an Explanation for Loss of Plasticitycited by 26
  8. 2024A Study of Plasticity Loss in On-Policy Deep Reinforcement Learningcited by 26
  9. 2023Symmetry Induces Structure and Constraint of Learningcited by 25
  10. 2024Plasticity Loss in Deep Reinforcement Learning: A Surveycited by 24
  11. 2025Mitigating Plasticity Loss in Continual Reinforcement Learning by Reducing Churncited by 22
  12. 2024Scalable Reinforcement Post-Training Beyond Static Human Prompts: Evolving Alignment via Asymmetric Self-Playcited by 21
30 more
  1. 2024Improving Deep Reinforcement Learning by Reducing the Chain Effect of Value and Policy Churncited by 16
  2. 2025Lightweight Online Adaption for Time Series Foundation Model Forecastscited by 14
  3. 2024Self-Normalized Resets for Plasticity in Continual Learningcited by 11
  4. 2024Remove Symmetries to Control Model Expressivity and Improve Optimizationcited by 8
  5. 2025Activation by Interval-wise Dropout: A Simple Way to Prevent Neural Networks from Plasticity Losscited by 6
  6. 2025Merge then Realign: Simple and Effective Modality-Incremental Continual Learning for Multimodal LLMscited by 6
  7. 2025Reinitializing weights vs units for maintaining plasticity in neural networkscited by 6
  8. 2024The Need for a Big World Simulator: A Scientific Challenge for Continual Learningcited by 5
  9. 2025Barriers for Learning in an Evolving World: Mathematical Understanding of Loss of Plasticitycited by 5
  10. 2026FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Controlcited by 5
  11. 2026Learning the Context of Errors: Black-Box Online Adaptation of Time Series Foundation Modelscited by 5
  12. 2024Efficient Language-instructed Skill Acquisition via Reward-Policy Co-Evolutioncited by 4
  13. 2026Predicting Plasticity in Deep Continual Learning: A Theoretical Perspectivecited by 4
  14. 2024Mitigating the Stability-Plasticity Dilemma in Adaptive Train Scheduling with Curriculum-Driven Continual DQN Expansioncited by 2
  15. 2025Sparsity-Driven Plasticity in Multi-Task Reinforcement Learningcited by 2
  16. 2023Towards Robust Continual Learning with Bayesian Adaptive Moment Regularizationcited by 1
  17. 2025Revisiting Clustering of Neural Bandits: Selective Reinitialization for Mitigating Loss of Plasticitycited by 1
  18. 2026Continual Policy Consolidation for Lifelong Robot Learningcited by 1
  19. 2026Convergent World Representations and Divergent Taskscited by 1
  20. 2026Plasticity-Enhanced Multi-Agent Mixture of Experts for Dynamic Objective Adaptation in UAVs-Assisted Emergency Communication Networkscited by 1
  21. 2026Can Scale Save Us From Plasticity Loss in Large Language Models?cited by 1
  22. 2026Portable Semantics, Private Dialects: Reuse and Negative Transfer in Latent Communication Between Language-Model Cellscited by 1
  23. 2024Forget Sharpness: Perturbed Forgetting of Model Biases Within SAM Dynamicscited by 0
  24. 2025Forgetting is Everywherecited by 0
  25. 2025Towards better dense rewards in Reinforcement Learning Applicationscited by 0
  26. 2026Beyond Attention: True Adaptive World Models via Spherical Kernel Operatorcited by 0
  27. 2026Adapt or Forget: Provable Tradeoffs Between Adam and SGD in Nonstationary Optimizationcited by 0
  28. 2026Attribution-Based Neuron Utility for Plasticity Restoration in Deep Networkscited by 0
  29. 2026When is Warmstarting Effective for Scaling Language Models?cited by 0
  30. 2026RIZZ: Routing Interactions to Near Zero-Interference Zones for Continual Adaptation of Black-Box Agentscited by 0