mlsanityData to 7 Oct 2026

Reawakening knowledge: Anticipatory recovery from catastrophic interference via structured training

arXiv 2403.09613 ↗First posted 14 Mar 2024cs.LG cs.CLcited by 6

Not on the map. Why?

Abstract (arXiv)

We explore the training dynamics of neural networks in a structured non-IID setting where documents are presented cyclically in a fixed, repeated sequence. Typically, networks suffer from catastrophic interference when training on a sequence of documents; however, we discover a curious and remarkable property of LLMs finetuned sequentially in this setting: they exhibit anticipatory behavior, recovering from the forgetting on documents before encountering them again. This behavior occurs even though the documents are never presented in context together. The behavior emerges and becomes more robust as the architecture scales up its number of parameters. Through comprehensive experiments and visualizations, we demonstrate a new mechanism by which over-parametrized neural networks can recover from catastrophic interference and uncover new insights into training over-parameterized networks in cyclically structured environments.

Key facts. Cited by 6 papers in the map (data through 7 Oct 2026). Builds on 52 of its 75 references that are in the map. First posted on arXiv 14 Mar 2024.

Builds on

52 of 75 references.

References · 52

  1. 2017Attention Is All You Needcited by 59,932
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  3. 2020Language Models are Few-Shot Learnerscited by 32,199
  4. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  5. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,267
  6. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  7. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  8. 2015Distilling the Knowledge in a Neural Networkcited by 10,076
  9. 2022PaLM: Scaling Language Modeling with Pathwayscited by 5,304
  10. 2016Overcoming catastrophic forgetting in neural networkscited by 4,718
  11. 2016Pointer Sentinel Mixture Modelscited by 2,992
  12. 2022Training Compute-Optimal Large Language Modelscited by 2,927
38 more
  1. 2022Emergent Abilities of Large Language Modelscited by 2,483
  2. 2016Learning without Forgettingcited by 2,465
  3. 2020The Pile: An 800GB Dataset of Diverse Text for Language Modelingcited by 2,346
  4. 2016iCaRL: Incremental Classifier and Representation Learningcited by 2,313
  5. 2022Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language modelscited by 1,970
  6. 2023Pythia: A Suite for Analyzing Large Language Models Across Training and Scalingcited by 1,734
  7. 2017Memory Aware Synapses: Learning what (not) to forgetcited by 1,094
  8. 2018Experience Replay for Continual Learningcited by 1,074
  9. 2020Dark Experience for General Continual Learning: a Strong, Simple Baselinecited by 895
  10. 2016Abstractive Text Summarization Using Sequence-to-Sequence RNNs and Beyondcited by 809
  11. 2022Quantifying Memorization Across Neural Language Modelscited by 755
  12. 2017Lifelong Learning with Dynamically Expandable Networkscited by 689
  13. 2018Overcoming catastrophic forgetting with hard attention to the taskcited by 672
  14. 2022Predictability and Surprise in Large Generative Modelscited by 209
  15. 2023Progressive Prompts: Continual Learning for Language Modelscited by 195
  16. 2023Continual Pre-training of Language Modelscited by 194
  17. 2019Meta-Learning Representations for Continual Learningcited by 189
  18. 2022Fine-tuned Language Models are Continual Learnerscited by 152
  19. 2021Self-Supervised Models are Continual Learnerscited by 149
  20. 2019Online Meta-Learningcited by 137
  21. 2020Random Reshuffling: Simple Analysis with Vast Improvementscited by 112
  22. 2021LFPT5: A Unified Framework for Lifelong Few-shot Language Learning Based on Prompt Tuning of T5cited by 103
  23. 2023To Repeat or Not To Repeat: Insights from Scaling LLM under Token-Crisiscited by 88
  24. 2020Continual Learning in Task-Oriented Dialogue Systemscited by 82
  25. 2021Online Continual Learning with Natural Distribution Shifts: An Empirical Study with Visual Datacited by 67
  26. 2019How Good is SGD with Random Shuffling?cited by 63
  27. 2020SGD with shuffling: optimal rates without component convexity and large epoch requirementscited by 57
  28. 2022A Simple Baseline that Questions the Use of Pretrained-Models in Continual Learningcited by 54
  29. 2019Learning-to-Learn Stochastic Gradient Descent with Biased Regularizationcited by 49
  30. 2020Online Continual Learning under Extreme Memory Constraintscited by 48
  31. 2022Datasheet for the Pilecited by 41
  32. 2022NISPA: Neuro-Inspired Stability-Plasticity Adaptation for Continual Learning in Sparse Networkscited by 38
  33. 2021Wanderlust: Online Continual Object Detection in the Real Worldcited by 34
  34. 2015Convergence Rate of Incremental Gradient and Newton Methodscited by 26
  35. 2020Wandering Within a World: Online Contextualized Few-Shot Learningcited by 22
  36. 2022Do Pre-trained Models Benefit Equally in Continual Learning?cited by 20
  37. 2019Sequential mastery of multiple visual tasks: Networks naturally learn to learn and forget to forgetcited by 13
  38. 2023Heterogeneous Continual Learningcited by 6

50 most cited of 52 All 52 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 2; 2025: 2; 2026: 2.

6 papers.

  1. 2024Continual Learning of Large Language Models: A Comprehensive Surveycited by 228
  2. 2024Avoiding Copyright Infringement via Large Language Model Unlearningcited by 22
  3. 2025From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularizationcited by 12
  4. 2025Fresh in memory: Training-order recency is linearly encoded in language model activationscited by 5
  5. 2026The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignmentcited by 1
  6. 2026Can a Language Model Learn Facts Continually in Its Weights?cited by 0