Mechanistic Interpretability

Follow this field Ask about this field

Recent momentum

emerging

0 papers in the last 28 days · 0.0% of indexed attention

Twelve weeks of publication activity for this topic as it is defined today.

Weekly history

Recent digests

What was published in this field, kept on the site without email delivery.

Period ending 2026-09-21

12 new papers

A weekly snapshot of new work published in Mechanistic Interpretability.

Period ending 2026-09-14

10 new papers

A weekly snapshot of new work published in Mechanistic Interpretability.

Period ending 2026-09-07

13 new papers

A weekly snapshot of new work published in Mechanistic Interpretability.

Inside this field

Focused directions

Focused topic

Model Interpretability

Within Mechanistic Interpretability

269 papersExplore

Focused topic

Mechanistic Interpretability

Within Mechanistic Interpretability

158 papersExplore

Focused topic

Interpretable Models

Within Mechanistic Interpretability

38 papersExplore

434 papers

Latest in Mechanistic Interpretability

Browse all topics →

InverseScope: Scalable Activation Inversion for Interpreting Large Language Models
Jun 9, 2025Yifan Luo, Zhennan Zhou, Bin DongModel Activations Model Interpretability
Kronecker Factorization Improves Efficiency and Interpretability of Sparse Autoencoders
May 28, 2025Vadim Kurochkin, Yaroslav Aksenov, Daniil Laptev +2Model Activations Speech Encoder
Tensorization is a powerful but underexplored tool for compression and interpretability of neural networks
May 26, 2025Safa Hamreras, Sukhbinder Singh, Román OrúsTensor Networks Neural Network
Bayesian Deep Learning for Discrete Choice
May 23, 2025Daniel F. Villarraga, Ricardo A. DazianoDiscrete Choice Models Deep Learning
Experiments with Optimal Model Trees
Mar 17, 2025Sabino Francesco Roselli, Eibe FrankDecision Trees Interpretable Models
Using Mechanistic Interpretability to Craft Adversarial Attacks against Large Language Models
Mar 8, 2025Thomas Winninger, Boussad Addad, Katarzyna KapustaLanguage Model Evasion Attacks Mechanistic Interpretability
Compositional Concept-Based Neuron-Level Interpretability for Deep Reinforcement Learning
Feb 2, 2025Zeyu Jiang, Hai Huang, Xingquan ZuoDeep Reinforcement Learning Model Interpretability
Surrogate Modeling for Explainable Predictive Time Series Corrections
Dec 27, 2024Alfredo Lopez, Florian SobieczkyMultivariate Forecasting Surrogate Models
MARS: A neurosymbolic approach for interpretable drug discovery
Oct 2, 2024Lauren Nicole DeLong, Yojana Gadiya, Paola Galdi +2Drug Discovery Neuro-Symbolic Framework
Interpretable Clustering: A Survey
Sep 1, 2024Lianyu Hu, Mudi Jiang, Junjie Dong +2Image Clustering Model Interpretability
Interpreting Global Perturbation Robustness of Image Models using Axiomatic Spectral Importance Decomposition
Aug 2, 2024Róisín Luo, James McDermott, Colm O'RiordanMechanistic Interpretability Perturbations
Debunking Grad-ECLIP: A Comprehensive Study on Its Incorrectness and Fundamental Principles for Model Interpretation
Date pendingYongjin Cui, Xiaohui FanInterpretable Models Systematic Study
SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?
Date pendingYuqiao Tan, Shizhu He, Jun Zhao +1Artificial Intelligence Agents Improving Sparse Autoencoders
Where is the Mind? Persona Vectors and LLM Individuation
Date pendingPierre Beckmann, Patrick ButlinPersonality Mechanistic Interpretability

Jun 9, 2025 · cs.LGJ/K move · Enter open · S save

InverseScope: Scalable Activation Inversion for Interpreting Large Language Models

Yifan Luo, Zhennan Zhou, Bin Dong

School of Mathematical Science, Peking University, Beijing, China · School of Science, Westlake University, Hangzhou, China · Beijing International Center for Mathematical Research and the New Cornerstone Science Laboratory, Peking University, Beijing, China · Center for Machine Learning Research, Peking University, Beijing, China

Understanding the internal representations of large language models (LLMs) is a central challenge in interpretability research. Existing feature interpretability methods often rely on strong structural assumptions--such as linearity or sparsity--that may not hold in practice. In this work, we introduce InverseScope, an assumption-light and scalable framework for interpreting neural activations via input inversion. Given a target activation, InverseScope characterizes its encoded information by generating natural-language inputs that produce nearby activations, grounding abstract internal states in concrete language. To overcome the prohibitive cost of sampling in high-dimensional activation spaces, we propose a novel control-layer conditioning architecture that substantially improves sample efficiency compared to prior token-prepending approaches. We demonstrate that InverseScope reveals rich geometric structure in LLM representation spaces, including sentence-level linear analogies. The framework scales to state-of-the-art open-source models of up to 14B parameters and generalizes to out-of-distribution inputs, enabling systematic analysis of activation neighborhoods.

Model Activations Model Interpretability Inverse Design

Open paper Ask about it PDF