Natural Language Processing

Execution-guided SQL generation enhances accuracy in text-to-SQL tasks by using execution results to select the most semantically consistent query, improving performance and reducing inference costs.

EAST: Prioritizing uncertainty in self-training refines reasoning of Large Language Models.

SEA: Stochastic Error Ascent efficiently discovers LLM knowledge gaps, outperforming existing methods in error detection with reduced cost.

Survey on efficient inference methods for Large Reasoning Models, focusing on mitigating token inefficiency while preserving quality.

ResearchBench: Benchmarking LLMs for Scientific Discovery via Inspiration-Based Task Decomposition.

ReFeed enhances multi-dimensional summarization by using reflective reasoning on feedback, mitigating trade-offs between dimensions and improving robustness.

ReaRAG enhances factuality in large reasoning models (LRMs) by integrating knowledge-guided reasoning with iterative retrieval augmented generation.

This survey presents a methodology-centered taxonomy of LLM agent systems, linking design principles to emergent behaviors and identifying future research directions.

OlymMATH: A new Olympiad-level math benchmark rigorously tests LLMs’ reasoning, revealing limitations and paving the way for advancements.

Survey on improving efficiency in large reasoning models across language, multimodality, and beyond.

Open Deep Search (ODS): Democratizing Search with Open-source Reasoning Agents.

MCTS-RAG: Combines Monte Carlo Tree Search with Retrieval-Augmented Generation to enhance small LMs’ reasoning on complex tasks.

Boost LLM reasoning by having models ‘Think Twice’! This novel method iteratively refines answers, significantly enhancing accuracy on complex tasks.

FFN Fusion: Parallelizing sequential computation in large language models for significant speedups!

LLMs falter on culturally adapted math problems, revealing a critical cultural bias.

V-SEEK accelerates LLM reasoning on open-hardware RISC-V platforms, achieving up to 3.0x speedup through optimized kernels and memory management.

This paper introduces deviation-factored post-training methods to enhance diversity and quality in creative LLM writing.

MARS: Optimizing prompts with multi-agent collaboration and Socratic learning for better LLM performance!

LEMMA: LLMs learn math via mistake analysis and correction, boosting performance without external critics.

XAttention: Antidiagonal scoring unlocks block-sparse attention, slashing compute costs in long-context Transformers without sacrificing accuracy.