A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Leslie Lamport
Introduces the Temporal Logic of Actions (TLA), a simple yet powerful logic for specifying and reasoning about concurrent systems, where systems and properties are expressed in the same logic.
Unknown
The paper identifies a hidden bias in Process Reward Models caused by severe imbalance in step-level feedback and proposes PRISM to reward the right reasoning.
Unknown
This paper shows that Process Reward Models can generalize from mathematical reasoning to code, enhancing LLM reasoning in test-time scaling across domains.
Unknown
This paper investigates process reward models for evaluating reasoning steps in thinking-with-images tasks, addressing the gap in text-centric PRM benchmarks.
Unknown
TechRAG proposes an evidence-gated multimodal agentic RAG architecture for technical literature reasoning, motivated by limitations of basic RAG.
Unknown
Proposes a multi-granularity context-enhanced RAG framework that retrieves over multimodal knowledge graphs to improve multimodal reasoning.
Unknown
The paper examines how reasoning can both help and hurt Vision-Language Models, revealing a dual nature where more thought sometimes reduces accuracy.
Ning Liu
Proposes using cross-model consensus among LLMs as a jury to select correct reasoning chains, outperforming process reward models for test-time scaling.
Unknown
OC-PRM introduces overcredit-contrastive training to reduce false positives in process reward models, improving step-level verification precision for reasoning LLMs.
Unknown
The paper proposes hierarchical process reward models as symbolic vision learners, supervising intermediate reasoning steps in vision tasks.
Unknown
Reasoning models fail to allocate test-time compute efficiently across multiple questions, revealing a gap between per-question and multi-question reasoning.
Unknown
The paper proposes an adaptive reasoning framework that learns when to allocate additional test-time compute, reducing over-computation in RL-trained reasoning models.