Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
50 篇论文 · 按点赞排序
Jiachen Zhu, Lingyu Yang, Rong Shan +6 authors
Researchers propose humanization capabilities for autonomous GUI agents to avoid detection by digital platforms, introducing a benchmark and methods to balance imitability with task performance.
Kangsan Kim, Minki Kang, Taeil Kim +3 authors
Memory transfer learning enables cross-domain code generation by leveraging unified memory pools, with performance improvements achieved through high-level abstraction rather than low-level code traces.
Yuqiao Tan, Minzheng Wang, Bo Liu +5 authors
PreRL applies reward-driven online updates to the marginal distribution in pre-train space, while DSRL uses NSR-PreRL to expand reasoning horizons before standard RL fine-tuning.
Hao Gao, Shaoyu Chen, Yifan Zhu +4 authors
A unified generator-discriminator framework for autonomous driving motion planning that improves stability and performance through diffusion-based trajectory generation and reinforcement learning optimization.
Sreyan Ghosh, Arushi Goel, Kaousheik Jayakumar +15 authors
Audio Flamingo Next represents a significant advancement in audio-language modeling with enhanced understanding capabilities, extended audio input lengths, and novel temporal reasoning mechanisms.
Tianyi Wang, Yixia Li, Long Li +6 authors
Sequence-Level PPO addresses instability in long-chain-of-thought reasoning by reformulating the process as a contextual bandit problem with decoupled value functions for improved efficiency.
Hippolyte Gisserot-Boukhlef, Nicolas Boizard, Emmanuel Malherbe +2 authors
Large language model evaluation faces challenges with rigid lexical methods that confuse problem-solving ability with formatting compliance, prompting the introduction of BERT-as-a-Judge for more robust, scalable assessment of generative outputs.
Shu Yang, Shenzhe Zhu, Hao Zhu +5 authors
Multi-user large language model agents face challenges in handling conflicting objectives, privacy preservation, and coordination efficiency in multi-principal decision-making scenarios.
Xuwei Ding, Skylar Zhai, Linxin Song +6 authors
Computer-use agents face significant safety vulnerabilities under unintended attack conditions where benign instructions lead to harmful outcomes through contextual or execution-based risks, with attack success rates exceeding 90% even in safety-aligned models.
Binbin Zheng, Xing Ma, Yiheng Liang +6 authors
SCOPE enhances on-policy distillation by adapting supervision paths based on trajectory correctness, using teacher-perplexity-weighted KL distillation for incorrect trajectories and student-perplexity-weighted MLE for correct ones, achieving superior reasoning performance.
Huacan Wang, Jie Zhou, Ningyan Zhu +8 authors
Sema Code presents an open AI coding framework that decouples the core agent engine from client interfaces, enabling shared reasoning capabilities across diverse development environments through a standalone npm library and modular architecture.
Roni Itkin, Noam Issachar, Yehonatan Keypur +3 authors
GlobalSplat introduces a global scene representation framework that achieves compact, consistent 3D Gaussian splatting with reduced computational overhead and improved inference speed.
Jiacheng Liu, Xiaohan Zhao, Xinyi Shang +1 authors
The study analyzes Claude Code's architecture, identifying five motivating human values and tracing them through thirteen design principles to specific implementation choices, including a core while-loop architecture and supporting systems for safety, context management, and extensibility.
Jaden Park, Jungtaek Kim, Jongwon Jeong +3 authors
Controllable environments with programmable exploration-exploitation balance are designed to evaluate language model agents' performance on embodied AI tasks, revealing distinct failure modes and demonstrating that reasoning models outperform other approaches.
Yifan Yu, Yuqing Jian, Junxiong Wang +12 authors
Introspective Diffusion Language Models address quality gaps with autoregressive models by enforcing introspective consistency through novel decoding algorithms and optimized inference engines.
Sahil Goyal, Swayam Agrawal, Gautham Govind Anil +3 authors
Elastic Looped Transformers utilize recurrent transformer architecture with weight-sharing and intra-loop self-distillation to achieve parameter-efficient visual generation with adjustable computational cost and generation quality.
Lifeng Chen, Tianqi You, Hao Liu +8 authors
ECHO is an efficient diffusion-based vision-language model for chest X-ray report generation that achieves faster inference through direct conditional distillation and response-asymmetric diffusion training while maintaining high clinical accuracy.
Charles Arnal, Vivien Cabannes, Taco Cohen +2 authors
Experience replay techniques for large language model post-training balance staleness variance and computational costs while maintaining performance and policy entropy.
Jean Kaddour
Target Policy Optimization separates policy update decisions from probability assignment in reinforcement learning, improving performance over standard policy gradient methods in sparse reward scenarios.
Ningyan Zhu, Huacan Wang, Jie Zhou +8 authors
OpenClaw's emergence in 2026 signifies a shift toward scalable personal AI agents requiring robust infrastructure for control and trustworthiness, addressed by SemaClaw's multi-agent framework with novel orchestration, safety, and context management components.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号