DeltaEvolve employs structured semantic delta and multi-level database to improve program evolution efficiency, reducing token use by 36.79%.

Jiachen Jiang, Tianyu Ding, Zhihui Zhu

InfMem enhances long-context reasoning accuracy by 10.17 points using the PreThink-Retrieve-Write protocol.

Xinyu Wang, Mingze Li, Peng Lu et al.

RLAnything employs closed-loop optimization to dynamically shape environment, policy, and reward models, boosting performance on tasks like OSWorld by 9.1%.

Yinjie Wang, Tianbao Xie, Ke Shen et al.

Reinforcement learning-based divide-and-conquer (DAC) training boosts LLM reasoning scalability, surpassing chain-of-thought (CoT) by 8.6%.

Xiao Liang, Zhong-Zhi Li, Zhenghao Lin et al.

TIC-VLA introduces a latency-aware semantic control interface, enabling robust robot navigation in dynamic environments despite multi-second reasoning delays.

Zhiyu Huang, Yun Zhang, Johnson Liu et al.

CHASE framework leverages pretrained protein language models and flow matching to generate high-fitness protein variants efficiently without external predictors.

Amaru Caceres Arroyo, Lea Bogensperger, Ahmed Allam et al.

UnifiedReward-Flex enhances vision generation with context-adaptive reasoning for personalized rewards.

Yibin Wang, Yuhang Zang, Feng Han et al.

SWE-Universe framework automates the construction of millions of real-world, verifiable software engineering environments using a custom-trained model and iterative validation.

Mouxiang Chen, Lei Zhang, Yunlong Feng et al.

Proposes PPTP, a layer-based decoupling method, to enhance neural network privacy without sacrificing generalization.

Xingli Fang, Jung-Eun Kim

Proposed RAG-based DziriBOT with DziriBERT achieves 88% accuracy on Algerian dialect understanding.

El Batoul Bechiri, Dihia Lanasri

HopFormer uses head-specific n-hop sparse masking for structure injection, avoiding positional encodings, with linear complexity.

Sanggeon Yun, Raheeb Hassan, Ryozo Masukawa et al.

Causal Forcing uses AR teacher for ODE initialization, bridging architecture gap, boosting real-time video quality by 19.3%.

Hongzhou Zhu, Min Zhao, Guande He et al.

Analyzed convergence rates of PCA for multiple probability measures under sparse and dense sampling, revealing a transition phenomenon; optimal in dense regime.

Gachon Erell, Jérémie Bigot, Elsa Cazelles

ECHO combines entropy and confidence-guided tree search with online pruning, significantly reducing collapse and bias in test-time RL, achieving +3.7% average gains on reasoning benchmarks.

Chu Zhao, Enneng Yang, Yuting Liu et al.

The paper reveals refusal behaviors in LLMs are controlled by multiple geometrically distinct directions, yet linear steering along any yields similar refusal effects.

Faaiz Joad, Majd Hawasly, Sabri Boughorbel et al.

Proposes daVinci-Agency, leveraging GitHub PR chains for long-horizon data synthesis; fine-tuning on 239 samples yields 47% improvement on Toolathlon.

Mohan Jiang, Dayuan Fu, Junhao Shi et al.

Proposes a multi-step reasoning framework with background stories to improve knowledge internalization and transfer in LLMs, boosting multi-hop reasoning by 10-20%.

Ya Gao, Kalle Kujanpää, Pekka Marttinen et al.

Analyzes representational failures in VLMs, proposes concept vector validation method.

Daniele Savietto, Declan Campbell, André Panisson et al.

Proposes xMemory, a hierarchical decoupling and aggregation method that improves agent memory retrieval, boosting answer quality and reducing redundancy by 15-20%.

Zhanghao Hu, Qinglin Zhu, Runcong Zhao et al.

ROMA plus GEPA+ lifts SEAL-0 accuracy by 9.9% over Kimi-Researcher.

Salaheddin Alzu'bi, Baran Nama, Arda Kaz et al.