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。
606 篇论文 · 按点赞排序
Abdelrahman Abouelenin, Atabak Ashfaq, Adam Atkinson +70 authors
Phi-4-Mini and Phi-4-Multimodal are compact yet highly capable language and multimodal models, featuring advanced attention mechanisms and LoRA adapters for efficient and powerful reasoning across various tasks.
Tiwei Bie, Maosong Cao, Kun Chen +28 authors
LLaDA2.0 converts auto-regressive models into discrete diffusion large language models with a novel training scheme, achieving superior performance and efficiency at scale.
Hasan Abed Al Kader Hammoud, Mohammad Zbeeb, Bernard Ghanem
Hala, a family of Arabic-centric instruction and translation models, achieves state-of-the-art results using a translate-and-tune pipeline, slerp merging, and fine-tuning on high-quality bilingual supervision.
Mingqi Wu, Zhihao Zhang, Qiaole Dong +9 authors
Research on enhancing reasoning capabilities in large language models using reinforcement learning reveals that accurate reward signals are crucial for performance improvement, and current benchmarks may be unreliable due to data contamination.
Abhay Kumar, Louis Owen, Nilabhra Roy Chowdhury +1 authors
ZClip is an adaptive gradient clipping algorithm that uses z-score-based anomaly detection to dynamically adjust clipping thresholds and prevent large gradient spikes during LLM training.
Fangfu Liu, Hanyang Wang, Yimo Cai +3 authors
Test-Time Scaling (TTS) in video generation improves video quality by adaptively sampling from noise space with feedback mechanisms, particularly demonstrated with the Tree-of-Frames method.
Zhaochen Su, Peng Xia, Hangyu Guo +12 authors
The survey outlines the evolution of multimodal AI from treating vision as static context to integrating it dynamically into the reasoning process, highlighting three stages and key challenges.
Yingzhe Peng, Gongrui Zhang, Miaosen Zhang +7 authors
A two-stage framework enhances reasoning in large multimodal models by first strengthening text-only reasoning and then generalizing it to multimodal tasks.
Jie Liu, Gongye Liu, Jiajun Liang +6 authors
Flow-GRPO combines online reinforcement learning with flow matching models through an ODE-to-SDE conversion and denoising reduction, improving sampling efficiency and performance across text-to-image tasks.
Fanqi Wan, Weizhou Shen, Shengyi Liao +7 authors
A framework called QwenLong-L1 enhances large reasoning models for long-context reasoning through reinforcement learning, achieving leading performance on document question-answering benchmarks.
Vishesh Tripathi, Tanmay Odapally, Indraneel Das +2 authors
A novel multimodal document chunking approach using Large Multimodal Models (LMMs) enhances RAG performance by accurately processing complex PDF documents, including multi-page tables and embedded visuals.
Hong Li, Houyuan Chen, Chongjie Ye +11 authors
Photometric stereo aims to recover high-quality surface normals under arbitrary lighting conditions, addressing challenges related to illumination-surface normal coupling and high-frequency geometric detail preservation.
Yangning Li, Weizhi Zhang, Yuyao Yang +17 authors
This survey integrates reasoning and retrieval in Large Language Models to improve factuality and multi-step inference, highlighting Synergized RAG-Reasoning frameworks and outlining future research directions.
Chen Hu, Haikuo Du, Heng Wang +64 authors
Step-DeepResearch, an end-to-end agent enhanced with a data synthesis strategy and progressive training, achieves expert-level capabilities in deep research scenarios, outperforming established models.
Daniil Moskovskiy, Nikita Sushko, Sergey Pletenev +2 authors
A pipeline generates multilingual synthetic detoxification data using LLMs, outperforming models trained on human-annotated datasets in few-shot settings.
Yue Liu, Hongcheng Gao, Shengfang Zhai +7 authors
GuardReasoner enhances LLM safety through enhanced reasoning capabilities using specialized training datasets and techniques, outperforming existing models on guardrail tasks.
Jianhao Yan, Yafu Li, Zican Hu +5 authors
LUFFY enhances zero-RL models with off-policy guidance, improving reasoning and generalization through balanced imitation and exploration.
Jiawei Gu, Yunzhuo Hao, Huichen Will Wang +5 authors
ThinkMorph, a unified model fine-tuned on interleaved reasoning traces, enhances multimodal reasoning by generating coherent text-image steps, achieving significant performance gains and demonstrating emergent capabilities.
Sara Vera Marjanović, Arkil Patel, Vaibhav Adlakha +14 authors
DeepSeek-R1 generates detailed reasoning chains, offering insights into model behavior and cognitive phenomena, with findings highlighting performance nuances and safety concerns.
Yi Peng, Chris, Xiaokun Wang +12 authors
Skywork R1V extends large language models to multimodal reasoning with efficient transfer, enhanced visual-text alignment, and dynamic reasoning chain optimization, achieving competitive performance in various benchmarks.
Shuming Ma, Hongyu Wang, Shaohan Huang +5 authors
BitNet b1.58 2B4T, a 1-bit Large Language Model with 2 billion parameters, matches the performance of full-precision models while improving computational efficiency.
Amirhosein Ghasemabadi, Di Niu
Large language models (LLMs) generate fluent and complex outputs but often fail to recognize their own mistakes and hallucinations. Existing approaches typically rely on external judges, multi-sample consistency, or text-based self-critique, which incur additional compute or correlate weakly with true correctness. We ask: can LLMs predict their own failures by inspecting internal states during inference? We introduce Gnosis, a lightweight self-awareness mechanism that enables frozen LLMs to perform intrinsic self-verification by decoding signals from hidden states and attention patterns. Gnosis passively observes internal traces, compresses them into fixed-budget descriptors, and predicts correctness with negligible inference cost, adding only ~5M parameters and operating independently of sequence length. Across math reasoning, open-domain question answering, and academic knowledge benchmarks, and over frozen backbones ranging from 1.7B to 20B parameters, Gnosis consistently outperforms strong internal baselines and large external judges in both accuracy and calibration. Moreover, it generalizes zero-shot to partial generations, enabling early detection of failing trajectories and compute-aware control. These results show that reliable correctness cues are intrinsic to generation process and can be extracted efficiently without external supervision.
Shijian Wang, Jiarui Jin, Xingjian Wang +6 authors
Video-Thinker, a multimodal large language model, autonomously reasons with videos using intrinsic grounding and captioning capabilities, achieving state-of-the-art performance on various video reasoning benchmarks.
Minjun Kim, Hyeonseok Lim, Hangyeol Yoo +10 authors
A large-scale investigation into constructing a fully open bilingual LLM for Korean using synthetic data demonstrates that such data can sustain pretraining and achieve performance comparable to multilingual baselines.
Kwanyoung Kim, Byeongsu Sim
PLADIS leverages sparse attention in cross-attention layers to enhance pre-trained text-to-image diffusion models, improving text alignment and human preference without additional training.
Hengyuan Xu, Wei Cheng, Peng Xing +8 authors
A diffusion-based model addresses copy-paste artifacts in text-to-image generation by using a large-scale paired dataset and a contrastive identity loss to balance identity fidelity and variation.
Huan-ang Gao, Jiayi Geng, Wenyue Hua +24 authors
This survey reviews self-evolving agents, focusing on mechanisms for adaptation in components like models, memory, and architecture, and discusses challenges in safety, scalability, and co-evolutionary dynamics.
Ningning Wang, Xavier Hu, Pai Liu +11 authors
A study on the efficiency-effectiveness trade-off in LLM-driven agent systems identifies optimal agent framework design to reduce costs while maintaining performance.
Shuaijie She, Yu Bao, Yu Lu +7 authors
DuPO is a dual learning framework that generates annotation-free feedback using a generalized duality, enhancing performance across various tasks without relying on costly labels.
Ziyu Liu, Zeyi Sun, Yuhang Zang +5 authors
Visual Reinforcement Fine-Tuning (Visual-RFT) enhances Large Vision-Language Models (LVLMs) through reinforcement learning with visual perception verifiable rewards, achieving competitive performance in various visual and reasoning tasks with limited data.
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