Understanding Generalization and Forgetting in In-Context Continual Learning

• URL: http://arxiv.org/abs/2605.28705v1
• Date: Wed, 27 May 2026 16:31:51 GMT
• Title: Understanding Generalization and Forgetting in In-Context Continual Learning
• Abstract summary: In-context learning enables Large Language Models to adapt to new tasks via prompt-based reasoning alone.<n>Existing theories study ICL in single-task settings, while real-world prompts often contain sequences of heterogeneous tasks.<n>We propose the first theoretical framework for in-context continual learning, modeling how a pretrained Transformer processes multiple sequential tasks within a single prompt.
• Score: 20.151689821430043
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: In-context learning (ICL) derives its power from enabling Large Language Models to adapt to new tasks via prompt-based reasoning alone, entirely bypassing the need for parameter updates. Existing theories primarily study ICL in single-task settings, while real-world prompts often contain sequences of heterogeneous tasks, leaving a gap in understanding whether Large Language Models implicitly perform continual learning during inference. To bridge this gap, we propose the first theoretical framework for in-context continual learning, modeling how a pretrained Transformer processes multiple sequential tasks within a single prompt through shared attention mechanisms. Focusing on linear and masked linear self-attention, we derive error expressions for model predictions under sequential task prompts and analyze their generalization and forgetting behavior. Our results reveal that standard attention mechanisms inevitably induce intertask interference by uniformly or causally aggregating historical contexts, leading to systematic bias. We further provide a bias-variance-interference decomposition of prediction error, characterizing when historical in-context information yields positive transfer or provable negative transfer. This analysis exposes fundamental limits of attention-based continual inference and offers theoretical explanations for order sensitivity and performance degradation in long prompts.

Related papers

• Enhancing Multimodal In-Context Learning via Inductive-Deductive Reasoning [65.15766304205657]
  In-context learning (ICL) allows large models to adapt to tasks using a few examples, yet its extension to vision-language models (VLMs) remains fragile.<n>Our analysis reveals that the fundamental limitation lies in an inductive gap, models often produce correct answers from flawed reasoning.<n>We introduce a framework that restructures multimodal ICL as a principled inductive-deductive process.
  arXiv  Detail & Related papers  (2026-05-04T09:18:19Z)
• A Bayesian Perspective on the Role of Epistemic Uncertainty for Delayed Generalization in In-Context Learning [3.236735486944393]
  In-context learning enables transformers to adapt to new tasks from a few examples at inference time, while grokking highlights that this generalization can emerge abruptly only after prolonged training.<n>We study task generalization and grokking in in-context learning using a Bayesian perspective, asking what enables the delayed transition from memorization to generalization.
  arXiv  Detail & Related papers  (2026-04-14T08:25:07Z)
• Beyond the Prompt in Large Language Models: Comprehension, In-Context Learning, and Chain-of-Thought [15.598263332303612]
  Large Language Models (LLMs) have demonstrated remarkable proficiency across diverse tasks.<n>This study dives into the foundations of these observations by addressing three critical questions.
  arXiv  Detail & Related papers  (2026-02-16T10:57:26Z)
• On the Paradoxical Interference between Instruction-Following and Task Solving [50.75960598434753]
  Instruction following aims to align Large Language Models (LLMs) with human intent by specifying explicit constraints on how tasks should be performed.<n>We reveal a counterintuitive phenomenon: instruction following can paradoxically interfere with LLMs' task-solving capability.<n>We propose a metric, SUSTAINSCORE, to quantify the interference of instruction following with task solving.
  arXiv  Detail & Related papers  (2026-01-29T17:48:56Z)
• Scaling Reasoning Hop Exposes Weaknesses: Demystifying and Improving Hop Generalization in Large Language Models [66.36240676392502]
  Chain-of-thought (CoT) reasoning has become the standard paradigm for enabling Large Language Models (LLMs) to solve complex problems.<n>Recent studies reveal a sharp performance drop in reasoning hop generalization scenarios.<n>We propose test-time correction of reasoning, a lightweight intervention method that dynamically identifies and deactivates ep heads in the reasoning process.
  arXiv  Detail & Related papers  (2026-01-29T03:24:32Z)
• Priors in Time: Missing Inductive Biases for Language Model Interpretability [58.07412640266836]
  We show that Sparse Autoencoders impose priors that assume independence of concepts across time, implying stationarity.<n>We introduce a new interpretability objective -- Temporal Feature Analysis -- which possesses a temporal inductive bias to decompose representations at a given time into two parts.<n>Our results underscore the need for inductive biases that match the data in designing robust interpretability tools.
  arXiv  Detail & Related papers  (2025-11-03T18:43:48Z)
• In-Context Learning Is Provably Bayesian Inference: A Generalization Theory for Meta-Learning [51.56484100374058]
  We introduce a principled risk decomposition that separates the total ICL risk into two components: Bayes Gap and Posterior Variance.<n>For a uniform-attention Transformer, we derive a non-asymptotic upper bound on this gap, which explicitly clarifies the dependence on the number of pretraining prompts.<n>The Posterior Variance is a model-independent risk representing the intrinsic task uncertainty.
  arXiv  Detail & Related papers  (2025-10-13T03:42:31Z)
• Learning Linear Regression with Low-Rank Tasks in-Context [8.347662730632047]
  In-context learning (ICL) is a key building block of modern large language models.<n>We analyze a linear attention model trained on low-rank regression tasks.<n>We find that statistical fluctuations in finite pre-training data induce an implicit regularization.
  arXiv  Detail & Related papers  (2025-10-06T07:27:49Z)
• Next-token pretraining implies in-context learning [0.0]
  We show how models adapt to context when trained on token sequences, especially from non-ergodic sources.<n>Our information-theoretic framework precisely predicts these in-distribution ICL dynamics.<n>We also show that a model's in-context performance on any task is mathematically coupled to the ensemble of tasks seen in pretraining.
  arXiv  Detail & Related papers  (2025-05-23T21:00:18Z)
• In-Context Linear Regression Demystified: Training Dynamics and Mechanistic Interpretability of Multi-Head Softmax Attention [52.159541540613915]
  We study how multi-head softmax attention models are trained to perform in-context learning on linear data.<n>Our results reveal that in-context learning ability emerges from the trained transformer as an aggregated effect of its architecture and the underlying data distribution.
  arXiv  Detail & Related papers  (2025-03-17T02:00:49Z)
• The Right Time Matters: Data Arrangement Affects Zero-Shot Generalization in Instruction Tuning [86.19804569376333]
  We show that zero-shot generalization happens very early during instruction tuning.<n>We propose a more grounded training data arrangement framework, Test-centric Multi-turn Arrangement.
  arXiv  Detail & Related papers  (2024-06-17T16:40:21Z)
• The mechanistic basis of data dependence and abrupt learning in an in-context classification task [0.3626013617212666]
  We show that specific distributional properties inherent in language control the trade-off or simultaneous appearance of two forms of learning. In-context learning is driven by the abrupt emergence of an induction head, which subsequently competes with in-weights learning. We propose that the sharp transitions in attention-based networks arise due to a specific chain of multi-layer operations necessary to achieve ICL.
  arXiv  Detail & Related papers  (2023-12-03T20:53:41Z)
• In-context Learning Generalizes, But Not Always Robustly: The Case of Syntax [36.98247762224868]
  In-context learning (ICL) is now a common method for teaching large language models (LLMs) new tasks. Do models infer the underlying structure of the task defined by the context, or do they rely on superficial generalizations that only generalize to identically distributed examples? In experiments with models from the GPT, PaLM, and Llama 2 families, we find large variance across LMs. The variance is explained more by the composition of the pre-training corpus and supervision methods than by model size.
  arXiv  Detail & Related papers  (2023-11-13T23:52:43Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.