Localized Definitions and Distributed Reasoning: A Proof-of-Concept Mechanistic Interpretability Study via Activation Patching

Nooshin Bahador

Home/Archive/arXiv

arXiv · 2504.02976

Localized Definitions and Distributed Reasoning: A Proof-of-Concept Mechanistic Interpretability Study via Activation Patching

Nooshin Bahador

View original source ↗Read source document ↗Export metadata ↓

Abstract

This study investigates the localization of knowledge representation in fine-tuned GPT-2 models using Causal Layer Attribution via Activation Patching (CLAP), a method that identifies critical neural layers responsible for correct answer generation. The model was fine-tuned on 9,958 PubMed abstracts (epilepsy: 20,595 mentions, EEG: 11,674 mentions, seizure: 13,921 mentions) using two configurations with validation loss monitoring for early stopping. CLAP involved (1) caching clean (correct answer) and corrupted (incorrect answer) activations, (2) computing logit difference to quantify model preference, and (3) patching corrupted activations with clean ones to assess recovery. Results revealed three findings: First, patching the first feedforward layer recovered 56% of correct preference, demonstrating that associative knowledge is distributed across multiple layers. Second, patching the final output layer completely restored accuracy (100% recovery), indicating that definitional knowledge is localised. The stronger clean logit difference for definitional questions further supports this localized representation. Third, minimal recovery from convolutional layer patching (13.6%) suggests low-level features contribute marginally to high-level reasoning. Statistical analysis confirmed significant layer-specific effects (p<0.01). These findings demonstrate that factual knowledge is more localized and associative knowledge depends on distributed representations. We also showed that editing efficacy depends on task type. Our findings not only reconcile conflicting observations about localization in model editing but also emphasize on using task-adaptive techniques for reliable, interpretable updates.

Explore related subjects

cs.LG cs.AI

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Nooshin Bahador. 2025-04-03. Localized Definitions and Distributed Reasoning: A Proof-of-Concept Mechanistic Interpretability Study via Activation Patching. https://arxiv.org/abs/2504.02976

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

CHAOSMINING: Benchmarking Post-Hoc Attribution with Sparse Informative Features in High Dimensions

Post-hoc attribution is widely used to identify important model inputs, but evaluating whether these attributions identify truly informative features is difficult because real datasets rarely provide reliable ground truth. We introduce a multimodal benchmark containing symbolic tabular, vision, and audio tasks with known informative feature sets. In the main benchmark conditions, informative variables, spatial regions, or channels occupy fixed input coordinates while the remaining inputs provide irrelevant or distracting information. We use the benchmark to study how attribution quality depends on predictive performance, irrelevant-feature burden and structure, model configuration, and attribution mechanism, while separately measuring identification, stability, and computational cost. In most symbolic-data sweeps, informative-set identification co-varies with predictive performance, while the relative ordering of attribution methods remains largely stable. Across modalities, no method dominates all architectures and conditions, and greater attribution complexity does not consistently improve identification. Simple gradient attribution is often competitive at lower computational cost, while the vision and audio results show that architecture and the form of irrelevant content materially affect attribution quality.

cs.LG↗

arXiv2026

HardCore Generation: Generating Hard UNSAT Problems for Data Augmentation

Efficiently determining the satisfiability of a boolean equation -- known as the SAT problem for brevity -- is crucial in various industrial problems. Recently, the advent of deep learning methods has introduced significant potential for enhancing SAT solving. However, a major barrier to the advancement of this field has been the scarcity of large, realistic datasets. The majority of current public datasets are either randomly generated or extremely limited, containing only a few examples from unrelated problem families. These datasets are inadequate for meaningful training of deep learning methods. In light of this, researchers have started exploring generative techniques to create data that more accurately reflect SAT problems encountered in practical situations. These methods have so far suffered from either the inability to produce challenging SAT problems or time-scalability obstacles. In this paper we address both by identifying and manipulating the key contributors to a problem's ``hardness'', known as cores. Although some previous work has addressed cores, the time costs are unacceptably high due to the expense of traditional heuristic core detection techniques. We introduce a fast core detection procedure that uses a graph neural network. Our empirical results demonstrate that we can efficiently generate problems that remain hard to solve and retain key attributes of the original example problems. We show via experiment that the generated synthetic SAT problems can be used in a data augmentation setting to provide improved prediction of solver runtimes.

cs.LG↗

arXiv2026

How Vulnerable Is My Learned Policy? Universal Adversarial Perturbation Attacks On Modern Behavior Cloning Policies

Imitation learning, also known as learning from demonstrations, is a popular approach to train AI models; however, the vulnerability of these models to adversarial attacks remains underexplored. We present the first systematic study of adversarial attacks, across a range of both classic and recently proposed imitation learning algorithms, including Vanilla Behavior Cloning (Vanilla BC), LSTM-GMM, Implicit Behavior Cloning (IBC), Diffusion Policy (DP), and Vector-Quantized Behavior Transformer (VQ-BET). We study the vulnerability of these methods to white-box, grey-box and black-box adversarial perturbations. Our experiments reveal that most existing methods are highly vulnerable to these attacks, including black-box transfer attacks that transfer across algorithms. White-box attacks cause at least a 65% reduction in average task success across all evaluated tasks and algorithms, while the black-box transfer attacks reduce task success by up to 88% on Lift, 99% on Can, and 100% on Square. To the best of our knowledge, we are the first to study and compare the vulnerabilities of different popular imitation learning algorithms to both white-box and black-box attacks. Our findings highlight the vulnerabilities of modern imitation learning algorithms, paving the way for future work in addressing such limitations. Videos and code are available at https://sites.google.com/view/uap-attacks-on-bc.

cs.LG↗