The Atlas of In-Context Learning: How Attention Heads Shape In-Context Retrieval Augmentation

• URL: http://arxiv.org/abs/2505.15807v1
• Date: Wed, 21 May 2025 17:59:01 GMT
• Title: The Atlas of In-Context Learning: How Attention Heads Shape In-Context Retrieval Augmentation
• Authors: Patrick Kahardipraja, Reduan Achtibat, Thomas Wiegand, Wojciech Samek, Sebastian Lapuschkin,
• Abstract summary: We shed light on the mechanism of in-context retrieval augmentation for question answering by viewing a prompt as a composition of informational components.<n>We propose an attribution-based method to identify specialized attention heads, revealing in-context heads that comprehend instructions and retrieve relevant contextual information.<n>We leverage the gained insights to trace the sources of knowledge used during inference, paving the way towards more safe and transparent language models.
• Score: 15.81218734180628
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Large language models are able to exploit in-context learning to access external knowledge beyond their training data through retrieval-augmentation. While promising, its inner workings remain unclear. In this work, we shed light on the mechanism of in-context retrieval augmentation for question answering by viewing a prompt as a composition of informational components. We propose an attribution-based method to identify specialized attention heads, revealing in-context heads that comprehend instructions and retrieve relevant contextual information, and parametric heads that store entities' relational knowledge. To better understand their roles, we extract function vectors and modify their attention weights to show how they can influence the answer generation process. Finally, we leverage the gained insights to trace the sources of knowledge used during inference, paving the way towards more safe and transparent language models.

Related papers

• Large Language Models are Limited in Out-of-Context Knowledge Reasoning [65.72847298578071]
  Large Language Models (LLMs) possess extensive knowledge and strong capabilities in performing in-context reasoning. This paper focuses on a significant aspect of out-of-context reasoning: Out-of-Context Knowledge Reasoning (OCKR), which is to combine multiple knowledge to infer new knowledge.
  arXiv  Detail & Related papers  (2024-06-11T15:58:59Z)
• Identifying Semantic Induction Heads to Understand In-Context Learning [103.00463655766066]
  We investigate whether attention heads encode two types of relationships between tokens present in natural languages. We find that certain attention heads exhibit a pattern where, when attending to head tokens, they recall tail tokens and increase the output logits of those tail tokens.
  arXiv  Detail & Related papers  (2024-02-20T14:43:39Z)
• Knowledge Graphs and Pre-trained Language Models enhanced Representation Learning for Conversational Recommender Systems [58.561904356651276]
  We introduce the Knowledge-Enhanced Entity Representation Learning (KERL) framework to improve the semantic understanding of entities for Conversational recommender systems. KERL uses a knowledge graph and a pre-trained language model to improve the semantic understanding of entities. KERL achieves state-of-the-art results in both recommendation and response generation tasks.
  arXiv  Detail & Related papers  (2023-12-18T06:41:23Z)
• Blending Reward Functions via Few Expert Demonstrations for Faithful and Accurate Knowledge-Grounded Dialogue Generation [22.38338205905379]
  We leverage reinforcement learning algorithms to overcome the above challenges by introducing a novel reward function. Our reward function combines an accuracy metric and a faithfulness metric to provide a balanced quality judgment of generated responses.
  arXiv  Detail & Related papers  (2023-11-02T02:42:41Z)
• Knowledge Graph Augmented Network Towards Multiview Representation Learning for Aspect-based Sentiment Analysis [96.53859361560505]
  We propose a knowledge graph augmented network (KGAN) to incorporate external knowledge with explicitly syntactic and contextual information. KGAN captures the sentiment feature representations from multiple perspectives, i.e., context-, syntax- and knowledge-based. Experiments on three popular ABSA benchmarks demonstrate the effectiveness and robustness of our KGAN.
  arXiv  Detail & Related papers  (2022-01-13T08:25:53Z)
• External Knowledge Augmented Text Visual Question Answering [0.6445605125467573]
  We propose a framework to extract, filter, and encode knowledge atop a standard multimodal transformer for vision language understanding tasks. We generate results comparable to the state-of-the-art on two publicly available datasets.
  arXiv  Detail & Related papers  (2021-08-22T13:21:58Z)
• Coarse-to-Careful: Seeking Semantic-related Knowledge for Open-domain Commonsense Question Answering [12.406729445165857]
  It is prevalent to utilize external knowledge to help machine answer questions that need background commonsense. We propose a semantic-driven knowledge-aware QA framework, which controls the knowledge injection in a coarse-to-careful fashion.
  arXiv  Detail & Related papers  (2021-07-04T10:56:36Z)
• Contextualized Knowledge-aware Attentive Neural Network: Enhancing Answer Selection with Knowledge [77.77684299758494]
  We extensively investigate approaches to enhancing the answer selection model with external knowledge from knowledge graph (KG) First, we present a context-knowledge interaction learning framework, Knowledge-aware Neural Network (KNN), which learns the QA sentence representations by considering a tight interaction with the external knowledge from KG and the textual information. To handle the diversity and complexity of KG information, we propose a Contextualized Knowledge-aware Attentive Neural Network (CKANN), which improves the knowledge representation learning with structure information via a customized Graph Convolutional Network (GCN) and comprehensively learns context-based and knowledge-based sentence representation via
  arXiv  Detail & Related papers  (2021-04-12T05:52:20Z)
• Multi-turn Dialogue Reading Comprehension with Pivot Turns and Knowledge [43.352833140317486]
  Multi-turn dialogue reading comprehension aims to teach machines to read dialogue contexts and solve tasks such as response selection and answering questions. This work makes the first attempt to tackle the above two challenges by extracting substantially important turns as pivot utterances. We propose a pivot-oriented deep selection model (PoDS) on top of the Transformer-based language models for dialogue comprehension.
  arXiv  Detail & Related papers  (2021-02-10T15:00:12Z)
• Reasoning over Vision and Language: Exploring the Benefits of Supplemental Knowledge [59.87823082513752]
  This paper investigates the injection of knowledge from general-purpose knowledge bases (KBs) into vision-and-language transformers. We empirically study the relevance of various KBs to multiple tasks and benchmarks. The technique is model-agnostic and can expand the applicability of any vision-and-language transformer with minimal computational overhead.
  arXiv  Detail & Related papers  (2021-01-15T08:37:55Z)

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.