Imagen de Portada

Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Minegishi, Gouki, Furuta, Hiroki, Iwasawa, Yusuke, Matsuo, Yutaka
Formato:	Preprint
Publicado:	2025
Materias:	Computation and Language Artificial Intelligence Machine Learning
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Towards Empirical Interpretation of Internal Circuits and Properties in Grokked Transformers on Modular Polynomials
por: Furuta, Hiroki, et al.
Publicado: (2024)

Beyond Induction Heads: In-Context Meta Learning Induces Multi-Phase Circuit Emergence
por: Minegishi, Gouki, et al.
Publicado: (2025)

Understanding Emergent Misalignment via Feature Superposition Geometry
por: Minegishi, Gouki, et al.
Publicado: (2026)

Topology of Reasoning: Understanding Large Reasoning Models through Reasoning Graph Properties
por: Minegishi, Gouki, et al.
Publicado: (2025)

Zipping the Thought: When and How Compressed Reasoning Data Works in LLM Post-Training
por: Matsutani, Kohsei, et al.
Publicado: (2026)

Bridging Lottery Ticket and Grokking: Understanding Grokking from Inner Structure of Networks
por: Minegishi, Gouki, et al.
Publicado: (2023)

Safe Transformer: An Explicit Safety Bit For Interpretable And Controllable Alignment
por: Feng, Jingyuan, et al.
Publicado: (2026)

Language Models Do Hard Arithmetic Tasks Easily and Hardly Do Easy Arithmetic Tasks
por: Gambardella, Andrew, et al.
Publicado: (2024)

Inconsistent Tokenizations Cause Language Models to be Perplexed by Japanese Grammar
por: Gambardella, Andrew, et al.
Publicado: (2025)

Exposing Limitations of Language Model Agents in Sequential-Task Compositions on the Web
por: Furuta, Hiroki, et al.
Publicado: (2023)

Mechanism of Task-oriented Information Removal in In-context Learning
por: Cho, Hakaze, et al.
Publicado: (2025)

Semantic Token Clustering for Efficient Uncertainty Quantification in Large Language Models
por: Cao, Qi, et al.
Publicado: (2026)

Self-Harmony: Learning to Harmonize Self-Supervision and Self-Play in Test-Time Reinforcement Learning
por: Wang, Ru, et al.
Publicado: (2025)

A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis
por: Gur, Izzeddin, et al.
Publicado: (2023)

Evaluating Adversarial Robustness of Concept Representations in Sparse Autoencoders
por: Li, Aaron J., et al.
Publicado: (2025)

RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMs
por: Matsutani, Kohsei, et al.
Publicado: (2025)

Geometric-Averaged Preference Optimization for Soft Preference Labels
por: Furuta, Hiroki, et al.
Publicado: (2024)

Omanic: Towards Step-wise Evaluation of Multi-hop Reasoning in Large Language Models
por: Gu, Xiaojie, et al.
Publicado: (2026)

AbsTopK: Rethinking Sparse Autoencoders For Bidirectional Features
por: Zhu, Xudong, et al.
Publicado: (2025)

MedRECT: A Medical Reasoning Benchmark for Error Correction in Clinical Texts
por: Iwase, Naoto, et al.
Publicado: (2025)

Large Language Models as Theory of Mind Aware Generative Agents with Counterfactual Reflection
por: Yang, Bo, et al.
Publicado: (2025)

WorldPack: Compressed Memory Improves Spatial Consistency in Video World Modeling
por: Oshima, Yuta, et al.
Publicado: (2025)

Diversity-driven Data Selection for Language Model Tuning through Sparse Autoencoder
por: Yang, Xianjun, et al.
Publicado: (2025)

Sparse Autoencoder Features for Classifications and Transferability
por: Gallifant, Jack, et al.
Publicado: (2025)

Sparse Autoencoder Decomposition of Clinical Sequence Model Representations: Feature Complexity, Task Specialisation, and Mortality Prediction
por: Sainsbury, Chris, et al.
Publicado: (2026)

Incorporating Hierarchical Semantics in Sparse Autoencoder Architectures
por: Muchane, Mark, et al.
Publicado: (2025)

$\infty$-MoE: Generalizing Mixture of Experts to Infinite Experts
por: Takashiro, Shota, et al.
Publicado: (2026)

Sparse but Wrong: Incorrect L0 Leads to Incorrect Features in Sparse Autoencoders
por: Chanin, David, et al.
Publicado: (2025)

Jacobian Sparse Autoencoders: Sparsify Computations, Not Just Activations
por: Farnik, Lucy, et al.
Publicado: (2025)

Improving Steering Vectors by Targeting Sparse Autoencoder Features
por: Chalnev, Sviatoslav, et al.
Publicado: (2024)

Residual Koopman Spectral Profiling for Predicting and Preventing Transformer Training Instability
por: Kim, Bum Jun, et al.
Publicado: (2026)

Temporal Sparse Autoencoders: Leveraging the Sequential Nature of Language for Interpretability
por: Bhalla, Usha, et al.
Publicado: (2025)

Feature Hedging: Correlated Features Break Narrow Sparse Autoencoders
por: Chanin, David, et al.
Publicado: (2025)

SAEMark: Steering Personalized Multilingual LLM Watermarks with Sparse Autoencoders
por: Yu, Zhuohao, et al.
Publicado: (2025)

Which Programming Language and What Features at Pre-training Stage Affect Downstream Logical Inference Performance?
por: Uchiyama, Fumiya, et al.
Publicado: (2024)

Static Word Embeddings for Sentence Semantic Representation
por: Wada, Takashi, et al.
Publicado: (2025)

Beyond Input Activations: Identifying Influential Latents by Gradient Sparse Autoencoders
por: Shu, Dong, et al.
Publicado: (2025)

Steering LLMs? Actually, Sparse Autoencoders can outperform simple baselines
por: Jørgensen, Mikkel Godsk, et al.
Publicado: (2026)

Thinking While Listening: Fast-Slow Recurrence for Long-Horizon Sequential Modeling
por: Takashiro, Shota, et al.
Publicado: (2026)

C-voting: Confidence-Based Test-Time Voting without Explicit Energy Functions
por: Kubo, Kenji, et al.
Publicado: (2026)