SCALPEL: Selective Capability Ablation via Low-rank Parameter Editing for Large Language Model Interpretability Analysis

• URL: http://arxiv.org/abs/2601.07411v1
• Date: Mon, 12 Jan 2026 10:54:18 GMT
• Title: SCALPEL: Selective Capability Ablation via Low-rank Parameter Editing for Large Language Model Interpretability Analysis
• Abstract summary: Large language models excel across diverse domains, yet their deployment in high-stakes systems remains limited by incomplete understanding of their internal mechanisms.<n>We present SCALPEL, a framework representing capabilities as low-rank subspaces rather than discrete modules.<n>By training LoRA to reduce distinguishing correct from incorrect answers while preserving general language modeling quality, SCALPEL identifies low-rank representations responsible for particular capabilities while remaining disentangled from others.
• Score: 5.684482188223278
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models excel across diverse domains, yet their deployment in healthcare, legal systems, and autonomous decision-making remains limited by incomplete understanding of their internal mechanisms. As these models integrate into high-stakes systems, understanding how they encode capabilities has become fundamental to interpretability research. Traditional approaches identify important modules through gradient attribution or activation analysis, assuming specific capabilities map to specific components. However, this oversimplifies neural computation: modules may contribute to multiple capabilities simultaneously, while single capabilities may distribute across multiple modules. These coarse-grained analyses fail to capture fine-grained, distributed capability encoding. We present SCALPEL (Selective Capability Ablation via Low-rank Parameter Editing for Large language models), a framework representing capabilities as low-rank parameter subspaces rather than discrete modules. Our key insight is that capabilities can be characterized by low-rank modifications distributed across layers and modules, enabling precise capability removal without affecting others. By training LoRA adapters to reduce distinguishing correct from incorrect answers while preserving general language modeling quality, SCALPEL identifies low-rank representations responsible for particular capabilities while remaining disentangled from others. Experiments across diverse capability and linguistic tasks from BLiMP demonstrate that SCALPEL successfully removes target capabilities while preserving general capabilities, providing fine-grained insights into capability distribution across parameter space. Results reveal that capabilities exhibit low-rank structure and can be selectively ablated through targeted parameter-space interventions, offering nuanced understanding of capability encoding in LLMs.

Related papers

• Qwen-Scope: Turning Sparse Features into Development Tools for Large Language Models [80.45129499188461]
  We introduce Qwen-Scope, an open-source suite of sparse autoencoders (SAEs) built on the Qwen model family.<n>We show that SAEs can go beyond post-hoc analysis to serve as practical interfaces for model development along four directions.
  arXiv  Detail & Related papers  (2026-05-12T10:01:06Z)
• 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)
• Compressed Sensing for Capability Localization in Large Language Models [58.574245030744834]
  Large language models (LLMs) exhibit a wide range of capabilities, including mathematical reasoning, code generation, and linguistic behaviors.<n>We show that many capabilities are highly localized to small subsets of attention heads within Transformer architectures.
  arXiv  Detail & Related papers  (2026-02-11T16:06:59Z)
• Control Reinforcement Learning: Interpretable Token-Level Steering of LLMs via Sparse Autoencoder Features [1.5874067490843806]
  Control Reinforcement Learning trains a policy to select SAE features for steering at each token, producing interpretable intervention logs.<n> Adaptive Feature Masking encourages diverse feature discovery while preserving singlefeature interpretability.<n>On Gemma 2 2B across MMLU, BBQ, GSM8K, HarmBench, and XSTest, CRL achieves improvements while providing per-token intervention logs.
  arXiv  Detail & Related papers  (2026-02-11T02:28:49Z)
• Finding the Translation Switch: Discovering and Exploiting the Task-Initiation Features in LLMs [69.28193153685893]
  Large Language Models (LLMs) frequently exhibit strong translation abilities, even without task-specific fine-tuning.<n>To demystify this process, we leverage Sparse Autoencoders (SAEs) and introduce a novel framework for identifying task-specific features.<n>Our work not only decodes a core component of the translation mechanism in LLMs but also provides a blueprint for using internal model mechanism to create more robust and efficient models.
  arXiv  Detail & Related papers  (2026-01-16T06:29:07Z)
• Ability Transfer and Recovery via Modularized Parameters Localization [35.06445666199432]
  Large language models can be continually pre-trained or fine-tuned to improve performance in specific domains, languages, or skills.<n>We propose ACT, which localizes ability-relevant channels via activation differences and selectively transfers only the corresponding parameters.
  arXiv  Detail & Related papers  (2026-01-14T11:42:39Z)
• Mechanistic Knobs in LLMs: Retrieving and Steering High-Order Semantic Features via Sparse Autoencoders [8.188989044347595]
  We propose a Sparse Autoencoder-based framework for retrieving and steering semantically interpretable internal features.<n>Using the Big Five personality traits as a case study, we demonstrate that our method enables precise, bidirectional steering of model behavior.
  arXiv  Detail & Related papers  (2026-01-06T12:40:37Z)
• Unlocking the Forgery Detection Potential of Vanilla MLLMs: A Novel Training-Free Pipeline [5.740204096484276]
  We propose Foresee, a training-free MLLM-based pipeline tailored for image forgery analysis.<n>Foresee employs a type-prior-driven strategy and utilizes a Flexible Feature Detector module to handle copy-move manipulations.<n>Our approach simultaneously achieves superior localization accuracy and provides more comprehensive textual explanations.
  arXiv  Detail & Related papers  (2025-11-17T14:49:57Z)
• Graft: Integrating the Domain Knowledge via Efficient Parameter Synergy for MLLMs [56.76586846269894]
  Multimodal Large Language Models (MLLMs) have achieved success across various domains.<n>Despite its importance, the study of knowledge sharing among domain-specific MLLMs remains largely underexplored.<n>We propose a unified parameter integration framework that enables modular composition of expert capabilities.
  arXiv  Detail & Related papers  (2025-06-30T15:07:41Z)
• Do We Really Need GNNs with Explicit Structural Modeling? MLPs Suffice for Language Model Representations [50.45261187796993]
  Graph Neural Networks (GNNs) fail to fully utilize structural information, whereas Multi-Layer Perceptrons (MLPs) exhibit a surprising ability in structure-aware tasks.<n>This paper introduces a comprehensive probing framework from an information-theoretic perspective.
  arXiv  Detail & Related papers  (2025-06-26T18:10:28Z)
• Exploiting Mixture-of-Experts Redundancy Unlocks Multimodal Generative Abilities [69.26544016976396]
  We exploit the redundancy within Mixture-of-Experts (MoEs) as a source of additional capacity for learning a new modality.<n>We preserve the original language generation capabilities by applying low-rank adaptation exclusively to the tokens of the new modality.
  arXiv  Detail & Related papers  (2025-03-28T15:21:24Z)
• Inducing, Detecting and Characterising Neural Modules: A Pipeline for Functional Interpretability in Reinforcement Learning [1.597617022056624]
  We show how encouraging sparsity and locality in network weights leads to the emergence of functional modules in RL policy networks.<n>Applying these methods to 2D and 3D MiniGrid environments reveals the consistent emergence of distinct navigational modules for different axes.
  arXiv  Detail & Related papers  (2025-01-28T17:02:16Z)
• Tuning-Free Accountable Intervention for LLM Deployment -- A Metacognitive Approach [55.613461060997004]
  Large Language Models (LLMs) have catalyzed transformative advances across a spectrum of natural language processing tasks. We propose an innovative textitmetacognitive approach, dubbed textbfCLEAR, to equip LLMs with capabilities for self-aware error identification and correction.
  arXiv  Detail & Related papers  (2024-03-08T19:18:53Z)
• Can Large Language Models Learn Independent Causal Mechanisms? [9.274428418715347]
  Large Language Models (LLMs) fall short on the same tasks in uncommon settings or with distribution shifts. We show that causal models, that learn abstract variables and causal relationships, can demonstrate increased robustness against changes in the distribution.
  arXiv  Detail & Related papers  (2024-02-04T23:04:02Z)
• Examining Scaling and Transfer of Language Model Architectures for Machine Translation [51.69212730675345]
  Language models (LMs) process sequences in a single stack of layers, and encoder-decoder models (EncDec) utilize separate layer stacks for input and output processing. In machine translation, EncDec has long been the favoured approach, but with few studies investigating the performance of LMs.
  arXiv  Detail & Related papers  (2022-02-01T16:20:15Z)

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.