Sparse Feature Coactivation Reveals Composable Semantic Modules in Large Language Models

• URL: http://arxiv.org/abs/2506.18141v1
• Date: Sun, 22 Jun 2025 19:01:13 GMT
• Title: Sparse Feature Coactivation Reveals Composable Semantic Modules in Large Language Models
• Abstract summary: We identify semantically coherent, context-consistent network components in large language models.<n>We show that ablating semantic components for countries and relations changes model outputs in predictable ways.<n>Within relation components themselves, nodes from later layers tend to have a stronger causal impact on model outputs.
• Score: 12.194272268150398
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We identify semantically coherent, context-consistent network components in large language models (LLMs) using coactivation of sparse autoencoder (SAE) features collected from just a handful of prompts. Focusing on country-relation tasks, we show that ablating semantic components for countries and relations changes model outputs in predictable ways, while amplifying these components induces counterfactual responses. Notably, composing relation and country components yields compound counterfactual outputs. We find that, whereas most country components emerge from the very first layer, the more abstract relation components are concentrated in later layers. Furthermore, within relation components themselves, nodes from later layers tend to have a stronger causal impact on model outputs. Overall, these findings suggest a modular organization of knowledge within LLMs and advance methods for efficient, targeted model manipulation.

Related papers

• SERE: Structural Example Retrieval for Enhancing LLMs in Event Causality Identification [50.174050426976514]
  Event Causality Identification (ECI) requires models to determine whether a given pair of events in a context exhibits a causal relationship.<n>Large Language Models (LLMs) have demonstrated strong performance across various NLP tasks, but their effectiveness in ECI remains limited due to biases in causal reasoning.<n>We propose SERE, a structural example retrieval framework that leverages LLMs' few-shot learning capabilities.
  arXiv  Detail & Related papers  (2026-05-05T12:50:19Z)
• PCA-Seg: Revisiting Cost Aggregation for Open-Vocabulary Semantic and Part Segmentation [58.1914505657064]
  This paper proposes a simple yet effective parallel cost aggregation (PCA-Seg) paradigm to alleviate the challenge of knowledge interference between class-level semantics and spatial context.<n>Experiments on eight benchmarks show that each parallel block in PCA-Seg adds merely 0.35M parameters while achieving state-of-the-art OSPS performance.
  arXiv  Detail & Related papers  (2026-03-18T09:26:43Z)
• Navigating the Concept Space of Language Models [0.0]
  We present Concept Explorer, a scalable interactive system for post-hoc exploration of SAE features.<n>Our approach constructs a multi-resolution manifold over SAE feature embeddings, enabling progressive navigation from coarse concept clusters to fine-grained neighborhoods.<n>We demonstrate the utility of Concept Explorer on SAE features extracted from SmolLM2, where it reveals coherent high-level structure, meaningful subclusters, and distinctive rare concepts.
  arXiv  Detail & Related papers  (2026-03-06T20:52:50Z)
• Bridging Collaborative Filtering and Large Language Models with Dynamic Alignment, Multimodal Fusion and Evidence-grounded Explanations [1.3702600718499687]
  We develop an online adaptation mechanism that incorporates new user interactions through lightweight modules.<n>We create a unified representation that seamlessly combines collaborative signals with visual and audio features.<n>Our approach maintains the efficiency of frozen base models while adding minimal computational overhead, making it practical for real-world deployment.
  arXiv  Detail & Related papers  (2025-10-02T02:43:24Z)
• Componentization: Decomposing Monolithic LLM Responses into Manipulable Semantic Units [0.0]
  Large Language Models (LLMs) often produce monolithic text that is hard to edit in parts.<n>We present componentization, an approach that decomposes model outputs into modular, independently editable units.
  arXiv  Detail & Related papers  (2025-09-10T00:15:33Z)
• ReMeREC: Relation-aware and Multi-entity Referring Expression Comprehension [29.50623143244436]
  ReMeREC aims to localize specified entities or regions in an image based on natural language descriptions.<n>We first construct a relation-aware, multi-entity REC dataset called ReMeX.<n>We then propose ReMeREC, a novel framework that jointly leverages visual and textual cues to localize multiple entities.
  arXiv  Detail & Related papers  (2025-07-22T11:23:48Z)
• PartComposer: Learning and Composing Part-Level Concepts from Single-Image Examples [31.17687322390452]
  PartComposer is a framework for part-level concept learning from single-image examples.<n>We propose a dynamic data synthesis pipeline generating diverse part compositions to address one-shot data scarcity.
  arXiv  Detail & Related papers  (2025-06-03T15:43:28Z)
• The Unreasonable Effectiveness of Model Merging for Cross-Lingual Transfer in LLMs [54.59207567677249]
  Large language models (LLMs) still struggle across tasks outside of high-resource languages.<n>In this work, we investigate cross-lingual transfer to lower-resource languages where task-specific post-training data is scarce.
  arXiv  Detail & Related papers  (2025-05-23T20:28:31Z)
• Core Context Aware Transformers for Long Context Language Modeling [50.774702091154204]
  We propose a plug-and-play Core Context Aware (CCA) Attention for efficient long-context modeling.<n>Our method automatically focuses and strengthens core context while diminishing redundancy during the learning process.<n>Our method is able to replace the self-attention module in existing Large Language Models with minimal fine-tuning cost.
  arXiv  Detail & Related papers  (2024-12-17T01:54:08Z)
• Entity-Aware Self-Attention and Contextualized GCN for Enhanced Relation Extraction in Long Sentences [5.453850739960517]
  We propose a novel model, Entity-aware Self-attention Contextualized GCN (ESC-GCN), which efficiently incorporates syntactic structure of input sentences and semantic context of sequences. Our model achieves encouraging performance as compared to existing dependency-based and sequence-based models.
  arXiv  Detail & Related papers  (2024-09-15T10:50:51Z)
• Learning to Extract Structured Entities Using Language Models [52.281701191329]
  Recent advances in machine learning have significantly impacted the field of information extraction. We reformulate the task to be entity-centric, enabling the use of diverse metrics. We contribute to the field by introducing Structured Entity Extraction and proposing the Approximate Entity Set OverlaP metric.
  arXiv  Detail & Related papers  (2024-02-06T22:15:09Z)
• SAGE: Bridging Semantic and Actionable Parts for GEneralizable Manipulation of Articulated Objects [9.500480417077272]
  We propose a novel framework that bridges semantic and actionable parts of articulated objects to achieve generalizable manipulation under natural language instructions. A part-grounding module maps the semantic parts into so-called Generalizable Actionable Parts (GAParts), which inherently carry information about part motion. An interactive feedback module is incorporated to respond to failures, which closes the loop and increases the robustness of the overall framework.
  arXiv  Detail & Related papers  (2023-12-03T07:22:42Z)
• Multi-Grained Multimodal Interaction Network for Entity Linking [65.30260033700338]
  Multimodal entity linking task aims at resolving ambiguous mentions to a multimodal knowledge graph. We propose a novel Multi-GraIned Multimodal InteraCtion Network $textbf(MIMIC)$ framework for solving the MEL task.
  arXiv  Detail & Related papers  (2023-07-19T02:11:19Z)
• EvEntS ReaLM: Event Reasoning of Entity States via Language Models [24.077262847151232]
  Nominally, Large Language models (LLM) have been exposed to procedural knowledge about how objects interact, yet our benchmarking shows they fail to reason about the world. In particular, our results indicate that our prompting technique is especially useful for unseen attributes (out-of-domain) or when only limited data is available.
  arXiv  Detail & Related papers  (2022-11-10T07:48:01Z)
• Compositional Generalization in Grounded Language Learning via Induced Model Sparsity [81.38804205212425]
  We consider simple language-conditioned navigation problems in a grid world environment with disentangled observations. We design an agent that encourages sparse correlations between words in the instruction and attributes of objects, composing them together to find the goal. Our agent maintains a high level of performance on goals containing novel combinations of properties even when learning from a handful of demonstrations.
  arXiv  Detail & Related papers  (2022-07-06T08:46:27Z)
• TAGPRIME: A Unified Framework for Relational Structure Extraction [71.88926365652034]
  TAGPRIME is a sequence tagging model that appends priming words about the information of the given condition to the input text. With the self-attention mechanism in pre-trained language models, the priming words make the output contextualized representations contain more information about the given condition. Extensive experiments and analyses on three different tasks that cover ten datasets across five different languages demonstrate the generality and effectiveness of TAGPRIME.
  arXiv  Detail & Related papers  (2022-05-25T08:57:46Z)
• Improving Entity Linking through Semantic Reinforced Entity Embeddings [16.868791358905916]
  We propose a method to inject fine-grained semantic information into entity embeddings to reduce the distinctiveness and facilitate the learning of contextual commonality. Based on our entity embeddings, we achieved new sate-of-the-art performance on entity linking.
  arXiv  Detail & Related papers  (2021-06-16T00:27:56Z)
• Joint Extraction of Entity and Relation with Information Redundancy Elimination [0.0]
  We propose a joint extraction model to solve the problem of redundant information and overlapping relations of the entity and relation extraction model. This model can directly extract multiple pairs of related entities without generating unrelated information. We also propose a recurrent neural network named-LSTM that enhances the ability of recurrent units to model sentences.
  arXiv  Detail & Related papers  (2020-11-27T05:47:26Z)
• Neural Function Modules with Sparse Arguments: A Dynamic Approach to Integrating Information across Layers [84.57980167400513]
  Neural Function Modules (NFM) aims to introduce the same structural capability into deep learning. Most of the work in the context of feed-forward networks combining top-down and bottom-up feedback is limited to classification problems. The key contribution of our work is to combine attention, sparsity, top-down and bottom-up feedback, in a flexible algorithm.
  arXiv  Detail & Related papers  (2020-10-15T20:43:17Z)
• Referring Image Segmentation via Cross-Modal Progressive Comprehension [94.70482302324704]
  Referring image segmentation aims at segmenting the foreground masks of the entities that can well match the description given in the natural language expression. Previous approaches tackle this problem using implicit feature interaction and fusion between visual and linguistic modalities. We propose a Cross-Modal Progressive (CMPC) module and a Text-Guided Feature Exchange (TGFE) module to effectively address the challenging task.
  arXiv  Detail & Related papers  (2020-10-01T16:02:30Z)
• Multidirectional Associative Optimization of Function-Specific Word Representations [86.87082468226387]
  We present a neural framework for learning associations between interrelated groups of words. Our model induces a joint function-specific word vector space, where vectors of e.g. plausible SVO compositions lie close together. The model retains information about word group membership even in the joint space, and can thereby effectively be applied to a number of tasks reasoning over the SVO structure.
  arXiv  Detail & Related papers  (2020-05-11T17:07:20Z)
• Probing Linguistic Features of Sentence-Level Representations in Neural Relation Extraction [80.38130122127882]
  We introduce 14 probing tasks targeting linguistic properties relevant to neural relation extraction (RE) We use them to study representations learned by more than 40 different encoder architecture and linguistic feature combinations trained on two datasets. We find that the bias induced by the architecture and the inclusion of linguistic features are clearly expressed in the probing task performance.
  arXiv  Detail & Related papers  (2020-04-17T09:17:40Z)
• A Dependency Syntactic Knowledge Augmented Interactive Architecture for End-to-End Aspect-based Sentiment Analysis [73.74885246830611]
  We propose a novel dependency syntactic knowledge augmented interactive architecture with multi-task learning for end-to-end ABSA. This model is capable of fully exploiting the syntactic knowledge (dependency relations and types) by leveraging a well-designed Dependency Relation Embedded Graph Convolutional Network (DreGcn) Extensive experimental results on three benchmark datasets demonstrate the effectiveness of our approach.
  arXiv  Detail & Related papers  (2020-04-04T14:59:32Z)

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.