Contextualize-then-Aggregate: Circuits for In-Context Learning in Gemma-2 2B

• URL: http://arxiv.org/abs/2504.00132v1
• Date: Mon, 31 Mar 2025 18:33:55 GMT
• Title: Contextualize-then-Aggregate: Circuits for In-Context Learning in Gemma-2 2B
• Authors: Aleksandra Bakalova, Yana Veitsman, Xinting Huang, Michael Hahn,
• Abstract summary: In-Context Learning (ICL) is an intriguing ability of large language models (LLMs)<n>We use causal interventions to identify information flow in Gemma-2 2B for five naturalistic ICL tasks.<n>We find that the model infers task information using a two-step strategy we call contextualize-then-aggregate.
• Score: 46.99314622487279
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: In-Context Learning (ICL) is an intriguing ability of large language models (LLMs). Despite a substantial amount of work on its behavioral aspects and how it emerges in miniature setups, it remains unclear which mechanism assembles task information from the individual examples in a fewshot prompt. We use causal interventions to identify information flow in Gemma-2 2B for five naturalistic ICL tasks. We find that the model infers task information using a two-step strategy we call contextualize-then-aggregate: In the lower layers, the model builds up representations of individual fewshot examples, which are contextualized by preceding examples through connections between fewshot input and output tokens across the sequence. In the higher layers, these representations are aggregated to identify the task and prepare prediction of the next output. The importance of the contextualization step differs between tasks, and it may become more important in the presence of ambiguous examples. Overall, by providing rigorous causal analysis, our results shed light on the mechanisms through which ICL happens in language models.

Related papers

• Learning Task Representations from In-Context Learning [73.72066284711462]
  Large language models (LLMs) have demonstrated remarkable proficiency in in-context learning. We introduce an automated formulation for encoding task information in ICL prompts as a function of attention heads. We show that our method's effectiveness stems from aligning the distribution of the last hidden state with that of an optimally performing in-context-learned model.
  arXiv  Detail & Related papers  (2025-02-08T00:16:44Z)
• Boosting Short Text Classification with Multi-Source Information Exploration and Dual-Level Contrastive Learning [12.377363857246602]
  We propose a novel model named MI-DELIGHT for short text classification.<n>It first performs multi-source information exploration to alleviate the sparsity issues.<n>Then, the graph learning approach is adopted to learn the representation of short texts.
  arXiv  Detail & Related papers  (2025-01-16T00:26:15Z)
• PICLe: Pseudo-Annotations for In-Context Learning in Low-Resource Named Entity Detection [56.916656013563355]
  In-context learning (ICL) enables Large Language Models to perform tasks using few demonstrations. We propose PICLe, a framework for in-context learning with noisy, pseudo-annotated demonstrations. We evaluate PICLe on five biomedical NED datasets and show that, with zero human annotation, PICLe outperforms ICL in low-resource settings.
  arXiv  Detail & Related papers  (2024-12-16T16:09:35Z)
• Aggregation Artifacts in Subjective Tasks Collapse Large Language Models' Posteriors [74.04775677110179]
  In-context Learning (ICL) has become the primary method for performing natural language tasks with Large Language Models (LLMs) In this work, we examine whether this is the result of the aggregation used in corresponding datasets, where trying to combine low-agreement, disparate annotations might lead to annotation artifacts that create detrimental noise in the prompt. Our results indicate that aggregation is a confounding factor in the modeling of subjective tasks, and advocate focusing on modeling individuals instead.
  arXiv  Detail & Related papers  (2024-10-17T17:16:00Z)
• Implicit In-context Learning [37.0562059811099]
  We introduce Implicit In-context Learning (I2CL), an innovative paradigm that reduces the inference cost of ICL to that of zero-shot learning with minimal information loss.<n>I2CL achieves few-shot level performance at zero-shot inference cost, and it exhibits robustness against variations in demonstration examples.
  arXiv  Detail & Related papers  (2024-05-23T14:57:52Z)
• Dual Operating Modes of In-Context Learning [8.664657381613695]
  In-context learning (ICL) exhibits dual operating modes: task learning, and task retrieval. Recent theoretical work investigates various mathematical models to analyze ICL. We introduce a probabilistic model, with which one can explain the dual operating modes of ICL simultaneously.
  arXiv  Detail & Related papers  (2024-02-29T03:06:10Z)
• 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)
• Label Words are Anchors: An Information Flow Perspective for Understanding In-Context Learning [77.7070536959126]
  In-context learning (ICL) emerges as a promising capability of large language models (LLMs) In this paper, we investigate the working mechanism of ICL through an information flow lens. We introduce an anchor re-weighting method to improve ICL performance, a demonstration compression technique to expedite inference, and an analysis framework for diagnosing ICL errors in GPT2-XL.
  arXiv  Detail & Related papers  (2023-05-23T15:26:20Z)
• Rethinking the Role of Demonstrations: What Makes In-Context Learning Work? [112.72413411257662]
  Large language models (LMs) are able to in-context learn by conditioning on a few input-label pairs (demonstrations) and making predictions for new inputs. We show that ground truth demonstrations are in fact not required -- randomly replacing labels in the demonstrations barely hurts performance. We find that other aspects of the demonstrations are the key drivers of end task performance.
  arXiv  Detail & Related papers  (2022-02-25T17:25:19Z)
• Document-Level Event Role Filler Extraction using Multi-Granularity Contextualized Encoding [40.13163091122463]
  Event extraction is a difficult task since it requires a view of a larger context to determine which spans of text correspond to event role fillers. We first investigate how end-to-end neural sequence models perform on document-level role filler extraction. We show that our best system performs substantially better than prior work.
  arXiv  Detail & Related papers  (2020-05-13T20:42:17Z)

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.