A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Unknown
This paper introduces Action Chain-of-Thought (ACoT) for Vision-Language-Action models, decomposing action generation into a step-by-step reasoning process to improve manipulation performance and interpretability.
Maxime M'eloux, Silviu Maniu, Franccois Portet, et al.
This paper investigates whether mechanistic interpretability methods are identifiable, proposing a framework to assess the uniqueness and reliability of explanations derived from neural network internals.
Unknown
This paper proposes a mathematical philosophy of mechanistic interpretability, defining MI and identifying preconditions for its success.
Unknown
This paper surveys recent progress in mechanistic interpretability techniques applied to LLM alignment, covering methods from circuit discovery to feature visualization.
Unknown
MIB is a benchmark for evaluating mechanistic interpretability methods, aiming to establish lasting standards for measuring real improvements.
Unknown
A survey organizing mechanistic interpretability across three dimensions, outlining how it bridges theory and practice for understanding AI systems.
Daking Rai, Yilun Zhou, Shi Feng, et al.
This paper provides a practical review of mechanistic interpretability for transformer-based language models, synthesizing methods, applications, and challenges.
Unknown
This paper proposes a theoretical foundation for mechanistic interpretability, centering causality as the core principle underlying interpretability methods.
Unknown
This paper proposes an automated workflow for discovering circuits in neural networks, aiming to reduce manual effort in mechanistic interpretability.
Unknown
This paper surveys open problems in mechanistic interpretability, aiming to clarify goals, methods, and future directions for the field.
Unknown
This review explores mechanistic interpretability as a tool for AI safety, assessing its potential to prevent catastrophic outcomes.
Zohaib Salahuddin, Henry C. Woodruff, Avishek Chatterjee, et al.
This review categorizes nine interpretability methods for deep neural networks in medical image analysis, evaluates progress in explanation assessment, and provides guidelines and future directions for trustworthy AI.