Experience-Guided Adaptation of Inference-Time Reasoning Strategies

📅 Publication date: Nov 14 2025

📑 Paper

🔗 Code: N/A

📝 Denoscription:

Experience-Guided Reasoner dynamically generates and optimizes computational strategies at inference time, adapting to problems using accumulated experience and improving accuracy and efficiency.

MicroVQA++: High-Quality Microscopy Reasoning Dataset with Weakly Supervised Graphs for Multimodal Large Language Model

📅 Publication date: Nov 14 2025

📑 Paper

🔗 Code

Dynamic Reflections: Probing Video Representations with Text Alignment

📅 Publication date: Nov 4 2025

📑 Paper

🔗 Code: N/A

MarsRL: Advancing Multi-Agent Reasoning System via Reinforcement Learning with Agentic Pipeline Parallelism

📅 Publication date: Nov 14 2025

📑 Paper

🔗 Code

📝 Denoscription:

MarsRL enhances multi-agent reasoning systems by optimizing all agents jointly, improving accuracy in complex reasoning tasks.

Test-Time Spectrum-Aware Latent Steering for Zero-Shot Generalization in Vision-Language Models

📅 Publication date: Nov 12 2025

📑 Paper

🔗 Code: N/A

LiteAttention: A Temporal Sparse Attention for Diffusion Transformers

📅 Publication date: Nov 14 2025

📑 Paper

🔗 Code

📝 Denoscription:

LiteAttention exploits temporal coherence in diffusion attention to accelerate video generation without quality loss.

P1: Mastering Physics Olympiads with Reinforcement Learning

📅 Publication date: Nov 17 2025

📑 Paper PDF

🔗 Code Repository

INDIBATOR: Diverse and Fact-Grounded Individuality for Multi-Agent Debate in Molecular Discovery

📅 Publication Date: Published on Feb 2 2025

📑 Paper: https://arxiv.org/pdf/2602.01815

🔗 Code: N/A

📝 Denoscription:

Multi-agent systems for molecular discovery that use individualized scientist profiles based on publication and molecular history outperform traditional role-based approaches.

OVD: On-policy Verbal Distillation

📅 Publication Date: Published on Jan 29

📑 Paper: https://arxiv.org/pdf/2601.21968

🔗 Code: https://OVD.github.io

📝 Denoscription:

On-policy Verbal Distillation (OVD) enables efficient knowledge transfer from teacher to student models by replacing token-level probability matching with trajectory matching.

PISA: Piecewise Sparse Attention Is Wiser for Efficient Diffusion Transformers

📅 Publication Date: Published on Feb 1

📑 Paper: https://arxiv.org/pdf/2602.01077

🔗 Code: https://github.com/xie-lab-ml/piecewise-sparse-attention

📝 Denoscription:

PISA is a novel sparse attention method that improves diffusion transformer efficiency by approximating non-critical attention blocks instead of discarding them, achieving faster processing.

DiscoX: Benchmarking Discourse-Level Translation task in Expert Domains

📅 Publication date: Nov 14 2025

📑 Paper PDF

🔗 Code: N/A

📝 Denoscription:

A new benchmark DiscoX and evaluation system Metric-S are introduced to assess discourse-level and expert-level Chinese-English translation, highlighting the challenges in achieving professional-grade machine translation.

Agent-R1: Training Powerful LLM Agents with End-to-End Reinforcement Learning

📅 Publication date: Nov 18 2025

📑 Paper: https://arxiv.org/pdf/2511.14460.pdf

🔗 Code: https://github.com/0russwest0/Agent-R1

📝 Denoscription:

A new training framework for RL-based LLM Agents is introduced, extending MDP methodology and demonstrating effectiveness on Multihop QA tasks.

UnSAMv2: Self-Supervised Learning Enables Segment Anything at Any Granularity

📅 Publication date: Nov 17 2025

📑 Paper: https://arxiv.org/pdf/2511.13714.pdf

🔗 Code: https://github.com/yujunwei04/UnSAMv2

Virtual Width Networks

📅 Publication date: Nov 14 2025

📑 Paper: https://arxiv.org/pdf/2511.11238.pdf

🔗 Code: N/A

📝 Denoscription:

Virtual Width Networks (VWN) enhance model efficiency by expanding representational width without increasing computational cost, accelerating optimization and improving loss reduction.

Part-X-MLLM: Part-aware 3D Multimodal Large Language Model

📅 Publication date: Nov 17 2025

📑 Paper: https://arxiv.org/pdf/2511.13647.pdf

🔗 Code: https://github.com/AiEson/Part-X-MLLM

InstructVLA: Vision-Language-Action Instruction Tuning from
  Understanding to Manipulation

📅 Publication date: Jul 23 2025

📑 Paper PDF

🔗 Code: N/A

📝 Denoscription:

InstructVLA is an end-to-end vision-language-action model that enhances manipulation performance while preserving vision-language reasoning through multimodal training and mixture-of-experts adaptation.

Training Long-Context, Multi-Turn Software Engineering Agents with
  Reinforcement Learning

📅 Publication date: Aug 5 2025

📑 Paper PDF

🔗 Code: N/A

OlmoEarth: Stable Latent Image Modeling for Multimodal Earth Observation

📅 Publication date: Nov 17 2025

📑 Paper PDF

🔗 Code Repository

Influence Guided Sampling for Domain Adaptation of Text Retrievers

📅 Publication Date: Jan 29

📑 Paper: https://arxiv.org/pdf/2601.21759

🔗 Code: N/A

📝 Denoscription:

An reinforcement learning-based sampling framework adaptively reweights training datasets to improve embedding model performance while reducing GPU costs.

Kronos: A Foundation Model for the Language of Financial Markets

📅 Publication Date: Aug 2, 2025

📑 Paper: https://arxiv.org/pdf/2508.02739

🔗 Code: https://github.com/shiyu-coder/Kronos

📝 Denoscription:

Kronos is a novel foundation model for financial K-line data, employing a specialized tokenizer and autoregressive pre-training on a massive dataset. It significantly outperforms existing models in forecasting, volatility prediction, and generating synthetic financial data.