mlsanityData to 7 Oct 2026

Continual Learning via Local Module Composition

arXiv 2111.07736 ↗First posted 15 Nov 2021cs.LG cs.AIcited by 61

Not on the map. Why?

Abstract (arXiv)

Modularity is a compelling solution to continual learning (CL), the problem of modeling sequences of related tasks. Learning and then composing modules to solve different tasks provides an abstraction to address the principal challenges of CL including catastrophic forgetting, backward and forward transfer across tasks, and sub-linear model growth. We introduce local module composition (LMC), an approach to modular CL where each module is provided a local structural component that estimates a module's relevance to the input. Dynamic module composition is performed layer-wise based on local relevance scores. We demonstrate that agnosticity to task identities (IDs) arises from (local) structural learning that is module-specific as opposed to the task- and/or model-specific as in previous works, making LMC applicable to more CL settings compared to previous works. In addition, LMC also tracks statistics about the input distribution and adds new modules when outlier samples are detected. In the first set of experiments, LMC performs favorably compared to existing methods on the recent Continual Transfer-learning Benchmark without requiring task identities. In another study, we show that the locality of structural learning allows LMC to interpolate to related but unseen tasks (OOD), as well as to compose modular networks trained independently on different task sequences into a third modular network without any fine-tuning. Finally, in search for limitations of LMC we study it on more challenging sequences of 30 and 100 tasks, demonstrating that local module selection becomes much more challenging in presence of a large number of candidate modules. In this setting best performing LMC spawns much fewer modules compared to an oracle based baseline, however, it reaches a lower overall accuracy. The codebase is available under https://github.com/oleksost/LMC.

Key facts. Cited by 61 papers in the map (data through 7 Oct 2026). Builds on 78 of its 97 references that are in the map. First posted on arXiv 15 Nov 2021.

Builds on

78 of 97 references.

References · 78

  1. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  2. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  3. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,861
  4. 2016Overcoming catastrophic forgetting in neural networkscited by 4,718
  5. 2017Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithmscited by 4,283
  6. 2013Auto-Encoding Variational Bayescited by 3,980
  7. 2017Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layercited by 3,413
  8. 2016Learning without Forgettingcited by 2,465
  9. 2016iCaRL: Incremental Classifier and Representation Learningcited by 2,313
  10. 2015Variational Inference with Normalizing Flowscited by 2,161
  11. 2013Describing Textures in the Wildcited by 2,110
  12. 2017Gradient Episodic Memory for Continual Learningcited by 1,871
38 more
  1. 2016Progressive Neural Networkscited by 1,332
  2. 2017Memory Aware Synapses: Learning what (not) to forgetcited by 1,094
  3. 2018Experience Replay for Continual Learningcited by 1,074
  4. 2014NICE: Non-linear Independent Components Estimationcited by 1,052
  5. 2020In Search of Lost Domain Generalizationcited by 1,042
  6. 2018On First-Order Meta-Learning Algorithmscited by 1,003
  7. 2018Deep Anomaly Detection with Outlier Exposurecited by 939
  8. 2017Lifelong Learning with Dynamically Expandable Networkscited by 689
  9. 2018Overcoming catastrophic forgetting with hard attention to the taskcited by 672
  10. 2018Progress & Compress: A scalable framework for continual learningcited by 516
  11. 2019Generalizing from a Few Examples: A Survey on Few-Shot Learningcited by 462
  12. 2018Meta-Learning for Semi-Supervised Few-Shot Classificationcited by 456
  13. 2018Do Deep Generative Models Know What They Don't Know?cited by 443
  14. 2019Online Continual Learning with Maximally Interfered Retrievalcited by 376
  15. 2015Neural Module Networkscited by 375
  16. 2016Expert Gate: Lifelong Learning with a Network of Expertscited by 369
  17. 2018Invertible Residual Networkscited by 364
  18. 2017Variational Continual Learningcited by 335
  19. 2018How to train your MAMLcited by 306
  20. 2019On Tiny Episodic Memories in Continual Learningcited by 299
  21. 2020Towards Continual Reinforcement Learning: A Review and Perspectivescited by 268
  22. 2020Gradient Starvation: A Learning Proclivity in Neural Networkscited by 262
  23. 2018Learning Not to Learn: Training Deep Neural Networks with Biased Datacited by 255
  24. 2018Meta-Learning: A Surveycited by 254
  25. 2018Selective Experience Replay for Lifelong Learningcited by 239
  26. 2019Recurrent Independent Mechanismscited by 219
  27. 2019Meta-Learning Representations for Continual Learningcited by 189
  28. 2019Learning to Remember: A Synaptic Plasticity Driven Framework for Continual Learningcited by 181
  29. 2018Recasting Gradient-Based Meta-Learning as Hierarchical Bayescited by 180
  30. 2018Towards Robust Evaluations of Continual Learningcited by 148
  31. 2020A Neural Dirichlet Process Mixture Model for Task-Free Continual Learningcited by 133
  32. 2021Avalanche: an End-to-End Library for Continual Learningcited by 124
  33. 2019A continual learning survey: Defying forgetting in classification taskscited by 118
  34. 2017Learning Independent Causal Mechanismscited by 104
  35. 2020Are Neural Nets Modular? Inspecting Functional Modularity Through Differentiable Weight Maskscited by 97
  36. 2020Efficient Continual Learning with Modular Networks and Task-Driven Priorscited by 87
  37. 2020A Wholistic View of Continual Learning with Deep Neural Networks: Forgotten Lessons and the Bridge to Active and Open World Learningcited by 86
  38. 2016Learning to Compose Neural Networks for Question Answeringcited by 85

50 most cited of 78 All 78 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 16; 2023: 20; 2024: 12; 2025: 10; 2026: 3.

61 papers.

Sort:

By area:

  1. 2023A Comprehensive Survey of Continual Learning: Theory, Method and Applicationcited by 860
  2. 2023Class-Incremental Learning: A Surveycited by 209
  3. 2022A Model or 603 Exemplars: Towards Memory-Efficient Class-Incremental Learningcited by 152
  4. 2023Modular Deep Learningcited by 82
  5. 2024Towards Modular LLMs by Building and Reusing a Library of LoRAscited by 72
  6. 2022Continual Learning with Foundation Models: An Empirical Study of Latent Replaycited by 61
  7. 2024Self-Expansion of Pre-trained Models with Mixture of Adapters for Continual Learningcited by 53
  8. 2022Multi-Head Adapter Routing for Cross-Task Generalizationcited by 33
  9. 2023Learnability and Algorithm for Continual Learningcited by 28
  10. 2022Combining Modular Skills in Multitask Learningcited by 27
  11. 2022How to Reuse and Compose Knowledge for a Lifetime of Tasks: A Survey on Continual Learning and Functional Compositioncited by 27
  12. 2024CLAP4CLIP: Continual Learning with Probabilistic Finetuning for Vision-Language Modelscited by 25
38 more
  1. 2023Catastrophic Forgetting in Deep Learning: A Comprehensive Taxonomycited by 24
  2. 2024Fine-tuning Reinforcement Learning Models is Secretly a Forgetting Mitigation Problemcited by 23
  3. 2022CompoSuite: A Compositional Reinforcement Learning Benchmarkcited by 21
  4. 2023Embodied Lifelong Learning for Task and Motion Planningcited by 17
  5. 2022Maslow's Hammer for Catastrophic Forgetting: Node Re-Use vs Node Activationcited by 16
  6. 2022Beyond Supervised Continual Learning: a Reviewcited by 11
  7. 2023Task Difficulty Aware Parameter Allocation & Regularization for Lifelong Learningcited by 11
  8. 2023Continual Learning Under Language Shiftcited by 9
  9. 2022Entropy-based Stability-Plasticity for Lifelong Learningcited by 8
  10. 2024Integrating Present and Past in Unsupervised Continual Learningcited by 8
  11. 2022Continual Feature Selection: Spurious Features in Continual Learningcited by 7
  12. 2022SHELS: Exclusive Feature Sets for Novelty Detection and Continual Learning Without Class Boundariescited by 7
  13. 2023Egocentric Planning for Scalable Embodied Task Achievementcited by 7
  14. 2024GPT4Rec: Graph Prompt Tuning for Streaming Recommendationcited by 7
  15. 2024Continual Learning with Neuromorphic Computing: Foundations, Methods, and Emerging Applicationscited by 7
  16. 2025Order-Robust Class Incremental Learning: Graph-Driven Dynamic Similarity Groupingcited by 7
  17. 2022Self-Activating Neural Ensembles for Continual Reinforcement Learningcited by 5
  18. 2024Adaptive Intelligence: leveraging insights from adaptive behavior in animals to build flexible AI systemscited by 5
  19. 2023Visually Grounded Continual Language Learning with Selective Specializationcited by 4
  20. 2024Compete and Compose: Learning Independent Mechanisms for Modular World Modelscited by 4
  21. 2025Online time series prediction using feature adjustmentcited by 4
  22. 2022Learning Label Modular Prompts for Text Classification in the Wildcited by 3
  23. 2023Studying Generalization on Memory-Based Methods in Continual Learningcited by 3
  24. 2023T-SaS: Toward Shift-aware Dynamic Adaptation for Streaming Datacited by 3
  25. 2023Modularity in Deep Learning: A Surveycited by 3
  26. 2025Model Inversion with Layer-Specific Modeling and Alignment for Data-Free Continual Learningcited by 3
  27. 2023On Conditional and Compositional Language Model Differentiable Promptingcited by 2
  28. 2024RECAST: Reparameterized, Compact weight Adaptation for Sequential Taskscited by 2
  29. 2025The Lie of the Average: How Class Incremental Learning Evaluation Deceives You?cited by 2
  30. 2025Compositional meta-learning through probabilistic task inferencecited by 2
  31. 2022Toward Sustainable Continual Learning: Detection and Knowledge Repurposing of Similar Taskscited by 1
  32. 2023SketchOGD: Memory-Efficient Continual Learningcited by 1
  33. 2024Bridging Neuroscience and AI: Environmental Enrichment as a Model for Forward Knowledge Transfercited by 1
  34. 2024Learning Mamba as a Continual Learner: Meta-Learning Selective State Space Models for Continual Learningcited by 1
  35. 2025Improvements of Dark Experience Replay and Reservoir Sampling towards Better Balance between Consolidation and Plasticitycited by 1
  36. 2025The Bayesian Approach to Continual Learning: An Overviewcited by 1
  37. 2025Separating the what and how of compositional computation to enable reuse and continual learningcited by 1
  38. 2026Provable Effects of Data Replay in Continual Learning: A Feature Learning Perspectivecited by 1

50 most cited of 61 All 61 (JSON)