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
Changliang Xia, Chengyou Jia, Zhuohang Dang +1 authors
DenseDiT, a generative model-based approach, achieves superior performance in real-world dense prediction tasks using minimal training data compared to existing methods.
Zhijian Xu, Yilun Zhao, Manasi Patwardhan +2 authors
A benchmark named LimitGen evaluates LLMs' ability to identify limitations in scientific papers, using synthetic and human-written datasets, and enhances their feedback with literature retrieval.
Yukang Cao, Chenyang Si, Jinghao Wang +1 authors
FreeMorph is a tuning-free method for image morphing that uses guidance-aware spherical interpolation and step-oriented variation trend to achieve high-quality results faster than existing methods.
Vladislav Bargatin, Egor Chistov, Alexander Yakovenko +1 authors
MEMFOF is a memory-efficient multi-frame optical flow method that achieves state-of-the-art performance with reduced GPU memory usage for high-resolution inputs.
Samir Khaki, Xiuyu Li, Junxian Guo +7 authors
SparseLoRA reduces computational cost and speeds up fine-tuning of LLMs by dynamically selecting a sparse subset of weights for loss and gradient computation.
Jianyu Wang, Zhiqiang Hu, Lidong Bing
A novel prompt optimization framework, PromptQuine, improves LLM performance by pruning random demonstrations into effective prompts using evolutionary search in low-data regimes.
Yehui Tang, Xiaosong Li, Fangcheng Liu +19 authors
Mixture of Grouped Experts (MoGE) improves expert load balancing and execution efficiency for large language models, enhancing throughput and cost-to-performance on Ascend NPUs.
Magnus Dierking, Christopher E. Mower, Sarthak Das +10 authors
ARK is an open-source Python-first framework that integrates modern imitation-learning algorithms and seamless simulation-physical robot interactions to simplify robotics development and deployment.
Yifan Shen, Yuanzhe Liu, Jingyuan Zhu +6 authors
SpatialReasoner-R1, a vision-language reasoning model, uses Multi-Model Monte Carlo Tree Search and fine-grained Direct Preference Optimization to improve spatial reasoning, setting a new state-of-the-art on SPATIALRGPT-Bench.
Bingchen Zhao, Despoina Magka, Minqi Jiang +20 authors
An Automated LLM Speedrunning Benchmark evaluates AI agents' ability to reproduce scientific results by leveraging NanoGPT speedrun tasks, indicating that even recent reasoning LLMs struggle with re-implementing known improvements.
Matthieu Zimmer, Xiaotong Ji, Rasul Tutunov +3 authors
A new framework using self-generated goal-conditioned MDPs and Monte Carlo Tree Search improves automated theorem proving by generating and pursuing subgoals, achieving state-of-the-art results on PutnamBench.
Qize Yang, Shimin Yao, Weixuan Chen +7 authors
The paper addresses challenges in multimodal reasoning by implementing context and logical rewards to enhance global context understanding and prevent shortcut problems, using a reasoning omni-modal benchmark called IntentBench.
Weida Wang, Changyong He, Jin Zeng +1 authors
A ToF depth denoising network using motion-invariant graph fusion enhances temporal stability and spatial sharpness, achieving state-of-the-art performance on synthetic and real datasets.
Baiqiao Yin, Qineng Wang, Pingyue Zhang +11 authors
A new benchmark, MindCube, shows that VLMs can improve their understanding of unseen spaces by forming internal spatial representations and reasoning over them.
Ying Guo, Xi Liu, Cheng Zhen +2 authors
An autoregressive framework for real-time interactive head generation improves interaction realism through bidirectional learning and detailed conversational state understanding.
Benjamin Feuer, Lennart Purucker, Oussama Elachqar +1 authors
MARVIS, a training-free method, enhances small vision-language models to predict across various data modalities with high accuracy using latent embeddings and spatial reasoning.
Yulun Jiang, Yekun Chai, Maria Brbić +1 authors
MARBLE is a challenging multimodal reasoning benchmark that tests the step-by-step reasoning capabilities of multimodal language models across complex tasks, highlighting their limitations in perception and reasoning.
Jian Chen, Wenye Ma, Penghang Liu +6 authors
MusiXQA, a new dataset for evaluating MLLMs on music sheet understanding, reveals limitations and enables the development of Phi-3-MusiX, an improved MLLM for this task.
Yalun Dai, Yangyu Huang, Xin Zhang +6 authors
DELT, a paradigm for optimizing data organization in language model training, enhances performance through Data Scoring, Data Selection, and Data Ordering, with Learnability-Quality Scoring and Folding Ordering as key components.
Nicholas Edwards, Yukyung Lee, Yujun +4 authors
RExBench evaluates the capability of LLM agents to autonomously implement research extensions, finding that current agents require significant human guidance to succeed.
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