{"total":88,"rows":[{"id":"2503.17547","title":"Learning Multi-Level Features with Matryoshka Sparse Autoencoders","created":"2025-03-21","cited_by_count":143,"area":"mechinterp"},{"id":"2607.15495","title":"Verbalizable Representations Form a Global Workspace in Language Models","created":"2026-07-16","cited_by_count":75,"area":null},{"id":"2503.01822","title":"Projecting Assumptions: The Duality Between Sparse Autoencoders and Concept Geometry","created":"2025-03-03","cited_by_count":61,"area":"mechinterp"},{"id":"2505.20322","title":"Beyond Prompt Engineering: Robust Behavior Control in LLMs via Steering Target Atoms","created":"2025-05-23","cited_by_count":24,"area":null},{"id":"2506.15679","title":"Dense SAE Latents Are Features, Not Bugs","created":"2025-06-18","cited_by_count":17,"area":"mechinterp"},{"id":"2506.20790","title":"Stochastic Parameter Decomposition","created":"2025-06-25","cited_by_count":16,"area":null},{"id":"2503.03601","title":"Feature-Level Insights into Artificial Text Detection with Sparse Autoencoders","created":"2025-03-05","cited_by_count":15,"area":"mechinterp"},{"id":"2505.20254","title":"Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEs","created":"2025-05-26","cited_by_count":14,"area":"mechinterp"},{"id":"2602.14111","title":"Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines?","created":"2026-02-15","cited_by_count":14,"area":"mechinterp"},{"id":"2601.05679","title":"Do Sparse Autoencoders Identify Reasoning Features in Language Models?","created":"2026-01-09","cited_by_count":12,"area":"lr"},{"id":"2506.19382","title":"Measuring and Guiding Monosemanticity","created":"2025-06-24","cited_by_count":10,"area":"mechinterp"},{"id":"2510.03659","title":"Does higher interpretability imply better utility? A Pairwise Analysis on Sparse Autoencoders","created":"2025-10-04","cited_by_count":10,"area":"mechinterp"},{"id":"2505.14536","title":"Breaking Bad Tokens: Detoxification of LLMs Using Sparse Autoencoders","created":"2025-05-20","cited_by_count":9,"area":"mechinterp"},{"id":"2503.24277","title":"Evaluating and Designing Sparse Autoencoders by Approximating Quasi-Orthogonality","created":"2025-03-31","cited_by_count":8,"area":"mechinterp"},{"id":"2506.00041","title":"Decoding Dense Embeddings: Sparse Autoencoders for Interpreting and Discretizing Dense Retrieval","created":"2025-05-28","cited_by_count":8,"area":"mechinterp"},{"id":"2507.08017","title":"Mechanistic Indicators of Understanding in Large Language Models","created":"2025-07-07","cited_by_count":8,"area":null},{"id":"2509.22033","title":"OrtSAE: Orthogonal Sparse Autoencoders Uncover Atomic Features","created":"2025-09-26","cited_by_count":8,"area":"mechinterp"},{"id":"2602.11881","title":"From Atoms to Trees: Building a Structured Feature Forest with Hierarchical Sparse Autoencoders","created":"2026-02-12","cited_by_count":6,"area":"mechinterp"},{"id":"2505.21364","title":"Towards Interpretability Without Sacrifice: Faithful Dense Layer Decomposition with Mixture of Decoders","created":"2025-05-27","cited_by_count":5,"area":null},{"id":"2507.12990","title":"Teach Old SAEs New Domain Tricks with Boosting","created":"2025-07-17","cited_by_count":5,"area":"mechinterp"},{"id":"2509.23717","title":"Measuring Sparse Autoencoder Feature Sensitivity","created":"2025-09-28","cited_by_count":4,"area":"mechinterp"},{"id":"2510.00468","title":"Feature Identification via the Empirical NTK","created":"2025-10-01","cited_by_count":4,"area":null},{"id":"2510.03186","title":"Superposition disentanglement of neural representations reveals hidden alignment","created":"2025-10-03","cited_by_count":4,"area":"mechinterp"},{"id":"2510.14073","title":"Exploratory Causal Inference in SAEnce","created":"2025-10-15","cited_by_count":4,"area":null},{"id":"2512.02004","title":"AlignSAE: Concept-Aligned Sparse Autoencoders","created":"2025-12-01","cited_by_count":4,"area":"mechinterp"},{"id":"2512.24975","title":"Attribution-Guided Distillation of Matryoshka Sparse Autoencoders","created":"2025-12-31","cited_by_count":4,"area":"mechinterp"},{"id":"2605.07922","title":"Tree SAE: Learning Hierarchical Feature Structures in Sparse Autoencoders","created":"2026-05-08","cited_by_count":4,"area":"mechinterp"},{"id":"2505.16077","title":"Ensembling Sparse Autoencoders","created":"2025-05-21","cited_by_count":3,"area":"mechinterp"},{"id":"2505.17769","title":"Inference-Time Decomposition of Activations (ITDA): A Scalable Approach to Interpreting Large Language Models","created":"2025-05-23","cited_by_count":3,"area":"mechinterp"},{"id":"2509.03738","title":"Mechanistic Interpretability with Sparse Autoencoder Neural Operators","created":"2025-09-03","cited_by_count":3,"area":"mechinterp"},{"id":"2602.05184","title":"Towards Worst-Case Guarantees with Scale-Aware Interpretability","created":"2026-02-05","cited_by_count":3,"area":null},{"id":"2504.20197","title":"Representation Learning on a Random Lattice","created":"2025-04-28","cited_by_count":2,"area":null},{"id":"2505.01372","title":"Evaluating Explanations: An Explanatory Virtues Framework for Mechanistic Interpretability -- The Strange Science Part I.ii","created":"2025-05-02","cited_by_count":2,"area":"mechinterp"},{"id":"2506.17673","title":"FaithfulSAE: Towards Capturing Faithful Features with Sparse Autoencoders without External Dataset Dependencies","created":"2025-06-21","cited_by_count":2,"area":"mechinterp"},{"id":"2511.06048","title":"Visual Exploration of Feature Relationships in Sparse Autoencoders with Curated Concepts","created":"2025-11-08","cited_by_count":2,"area":"mechinterp"},{"id":"2602.04718","title":"Superposition Without Interference? Towards Isolated Interventions via Almost Orthogonal Features in Language Models","created":"2026-02-04","cited_by_count":2,"area":"mechinterp"},{"id":"2604.14434","title":"Geometric Routing Enables Causal Expert Control in Mixture of Experts","created":"2026-04-15","cited_by_count":2,"area":"moe"},{"id":"2605.12122","title":"Disentangled Sparse Representations for Concept-Separated Diffusion Unlearning","created":"2026-05-12","cited_by_count":2,"area":"diffusion"},{"id":"2605.12874","title":"Descriptive Collision in Sparse Autoencoder Auto-Interpretability: When One Explanation Describes Many Features","created":"2026-05-13","cited_by_count":2,"area":"mechinterp"},{"id":"2608.19338","title":"Mechanistic Tomography: Designed Measurement for Control-Oriented Interpretability","created":"2026-08-19","cited_by_count":2,"area":"mechinterp"},{"id":"2511.17735","title":"Towards Open-Ended Visual Scientific Discovery with Sparse Autoencoders","created":"2025-11-21","cited_by_count":1,"area":"mechinterp"},{"id":"2512.02194","title":"Enforcing Orderedness to Improve Feature Consistency","created":"2025-12-01","cited_by_count":1,"area":"mechinterp"},{"id":"2602.13524","title":"Singular Vectors of Attention Heads Align with Features","created":"2026-02-13","cited_by_count":1,"area":"mechinterp"},{"id":"2604.24936","title":"A Unifying Framework for Unsupervised Concept Extraction","created":"2026-04-27","cited_by_count":1,"area":"mechinterp"},{"id":"2605.08012","title":"Position: Mechanistic Interpretability Must Disclose Identification Assumptions for Causal Claims","created":"2026-05-08","cited_by_count":1,"area":null},{"id":"2605.08891","title":"Bilinear autoencoders find interpretable manifolds","created":"2026-05-09","cited_by_count":1,"area":"mechinterp"},{"id":"2605.28870","title":"Representation Alignment Rests on Linear Structure","created":"2026-05-22","cited_by_count":1,"area":null},{"id":"2606.07007","title":"A Geometric View for Understanding Concept Learning and Neuron Interpretation in Sparse Autoencoders","created":"2026-06-05","cited_by_count":1,"area":"mechinterp"},{"id":"2606.12138","title":"Unstable Features, Reproducible Subspaces: Understanding Seed Dependence in Sparse Autoencoders","created":"2026-06-10","cited_by_count":1,"area":"mechinterp"},{"id":"2606.22994","title":"Do Sparse Autoencoders Learn Meaningful Concept Hierarchies?","created":"2026-06-22","cited_by_count":1,"area":"mechinterp"},{"id":"2607.08605","title":"When Structured Sparse Autoencoders Learn Consistent Concepts Across Modalities","created":"2026-07-09","cited_by_count":1,"area":"mechinterp"},{"id":"2607.16448","title":"Retrieval is Enough: Training-Free Interpretability with a Tool-Using Agent","created":"2026-07-17","cited_by_count":1,"area":"agents"},{"id":"2607.20596","title":"Are Single-Token Sparse Autoencoder Features Causally Necessary? Layer-Depth and SAE-Family Effects","created":"2026-07-22","cited_by_count":1,"area":"mechinterp"},{"id":"2608.07594","title":"Scaling Inherently Interpretable Language Models","created":"2026-08-06","cited_by_count":1,"area":"diffusion"},{"id":"2609.01936","title":"Sparse Readout Prism: Explaining Logit-Lens Scores in Features Instead of Tokens","created":"2026-09-01","cited_by_count":1,"area":null},{"id":"2609.06862","title":"Feature Superposition in Neural Networks: From Theory to Practice","created":"2026-09-06","cited_by_count":1,"area":"mechinterp"},{"id":"2609.19150","title":"Sampling Reveals Style: Unsupervised, Training-Free Discovery of Prompt-Conditional Stylistic Axes in LLM Activations","created":"2026-07-20","cited_by_count":1,"area":null},{"id":"2505.11836","title":"SplInterp: Improving our Understanding and Training of Sparse Autoencoders","created":"2025-05-17","cited_by_count":0,"area":"mechinterp"},{"id":"2602.00924","title":"Supervised sparse auto-encoders for interpretable and compositional representations","created":"2026-01-31","cited_by_count":0,"area":"mechinterp"},{"id":"2602.23826","title":"GLUScope: A Tool for Analyzing GLU Neurons in Transformer Language Models","created":"2026-02-27","cited_by_count":0,"area":null},{"id":"2604.00208","title":"Similarity of Neural Network Representations in Superposition","created":"2026-03-31","cited_by_count":0,"area":"mechinterp"},{"id":"2604.19083","title":"ProjLens: Unveiling the Role of Projectors in Multimodal Model Safety","created":"2026-04-21","cited_by_count":0,"area":"mllm"},{"id":"2605.03160","title":"Steering grids for sparse-autoencoder features: when a top-context label names an activation regime rather than a causal axis","created":"2026-05-04","cited_by_count":0,"area":"mechinterp"},{"id":"2605.05223","title":"Structural Instability of Feature Composition","created":"2026-04-18","cited_by_count":0,"area":"mechinterp"},{"id":"2605.28149","title":"Sign-Aware Gated Sparse Autoencoders: Modeling Anticorrelated Features with Bi-Jump-ReLU Activations","created":"2026-05-27","cited_by_count":0,"area":"mechinterp"},{"id":"2605.29384","title":"Latent Terms: Dense Retrievers Contain Trivially Extractable BM25-ready Zipfian Vocabularies","created":"2026-05-28","cited_by_count":0,"area":null},{"id":"2606.02385","title":"How Optimality Structures Sparse Dictionaries: Theory for Interpreting SAE Representations","created":"2026-06-01","cited_by_count":0,"area":"mechinterp"},{"id":"2606.14040","title":"Decompose Sparsely Where You Should, Absorb Densely Where You Should No","created":"2026-06-12","cited_by_count":0,"area":"mechinterp"},{"id":"2606.15054","title":"Size Doesn't Matter: Cosine-Scored Sparse Autoencoders","created":"2026-06-13","cited_by_count":0,"area":"mechinterp"},{"id":"2606.20347","title":"Critical Percolation as a Synthetic Data Model for Interpretability","created":"2026-06-18","cited_by_count":0,"area":null},{"id":"2606.26396","title":"At the Edge of Understanding: Sparse Autoencoders Trace The Limits of Transformer Generalization","created":"2026-06-24","cited_by_count":0,"area":"mechinterp"},{"id":"2606.32008","title":"Surrogate Fidelity: When Can Open LLMs Explain Closed Ones?","created":"2026-06-30","cited_by_count":0,"area":"mechinterp"},{"id":"2608.03507","title":"ChronoLens: Measuring Language Change Across Time, Languages, and Linguistic Levels","created":"2026-08-04","cited_by_count":0,"area":null},{"id":"2608.11197","title":"Beyond a Bag of Features: Set-Level Instability in Sparse Autoencoders","created":"2026-08-11","cited_by_count":0,"area":"mechinterp"},{"id":"2608.13337","title":"Where You Measure Decides What You Measure: Position Selection in Ablation-Based SAE Evaluation","created":"2026-08-13","cited_by_count":0,"area":"mechinterp"},{"id":"2608.20134","title":"Feature Evolution and Migration during Vision Transformer Training","created":"2026-08-20","cited_by_count":0,"area":"mechinterp"},{"id":"2608.27020","title":"Representation Measurements Under Function-Preserving Reparameterizations","created":"2026-08-27","cited_by_count":0,"area":null},{"id":"2608.28924","title":"Causal Interventions Reveal Typologically Organized Syntactic Mechanisms in Multilingual Language Models","created":"2026-08-28","cited_by_count":0,"area":null},{"id":"2609.09113","title":"SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?","created":"2026-09-08","cited_by_count":0,"area":"mechinterp"},{"id":"2609.18612","title":"Weakening Neurons: An Input-Output Functionality in Transformers with Outsize Influence","created":"2026-09-16","cited_by_count":0,"area":null},{"id":"2609.29328","title":"Grammatical \"grandmother neurons\" are rare in LLMs","created":"2026-09-24","cited_by_count":0,"area":null},{"id":"2609.29551","title":"HiPACE: Hierarchical Phase-Boundary Analysis and Controlled Evaluation of Feature Absorption in Sparse Autoencoders","created":"2026-08-26","cited_by_count":0,"area":"mechinterp"},{"id":"2609.34326","title":"Routing Without Embeddings: Fast And Interpretable Routing With Regular Expressions","created":"2026-09-28","cited_by_count":0,"area":"mechinterp"},{"id":"2609.36294","title":"When Trees Are Not Enough: Learning Mixed-Topology Feature Graphs with Adaptive Graph Sparse Autoencoders","created":"2026-09-28","cited_by_count":0,"area":"mechinterp"},{"id":"2609.37857","title":"Active Budget Can Kill Sensitivity: Diagnosing and Repairing TopK Sparse Autoencoder Reliability","created":"2026-09-29","cited_by_count":0,"area":"mechinterp"},{"id":"2610.04020","title":"ClasSAE: Class-Aligned Sparse Autoencoders via Differentiable Feature-Class Affinity","created":"2026-10-02","cited_by_count":0,"area":"mechinterp"},{"id":"2610.18155","title":"When Do Learned Blocks Preserve Causal Meaning? Block Identifiability and Constructive Causal Abstractions","created":"2026-10-13","cited_by_count":0,"area":null},{"id":"2610.19907","title":"Sparse Transformer: a Controlled Testbed for Mechanistic Interpretability","created":"2026-10-14","cited_by_count":0,"area":"mechinterp"}],"next_cursor":null}