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。
Lingrui Mei, Jiayu Yao, Yuyao Ge +12 authors
Context Engineering systematically optimizes information payloads for Large Language Models, addressing gaps in generating sophisticated, long-form outputs.
50 篇论文 · 按点赞排序
Zixiao Wang, Yuxin Wang, Xiaorui Wang +8 authors
MetaStone-S1, a reflective generative model using a self-supervised process reward model, achieves high performance with reduced parameters and supports test time scaling.
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.
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.
Luke Rivard, Sun Sun, Hongyu Guo +2 authors
NeuralOS, a neural framework combining RNNs and diffusion-based rendering, simulates realistic GUIs by predicting screen frames in response to user inputs.
Senqiao Yang, Junyi Li, Xin Lai +3 authors
VisionThink dynamically adjusts image resolution and visual token processing for efficient and effective vision-language tasks, improving performance and reducing computational cost.
Sangmin Bae, Yujin Kim, Reza Bayat +8 authors
Mixture-of-Recursions (MoR) combines parameter sharing and adaptive computation in a Recursive Transformer to improve efficiency, reduce memory usage, and enhance performance across different model scales.
Yana Wei, Liang Zhao, Jianjian Sun +15 authors
A two-stage paradigm using Qwen2.5-VL-7B and multimodal reinforcement learning achieves state-of-the-art performance in visual reasoning benchmarks.
Yifan Wang, Jianjun Zhou, Haoyi Zhu +7 authors
A permutation-equivariant neural network, $\pi^3$, reconstructs visual geometry without a fixed reference view, achieving state-of-the-art performance in camera pose estimation, depth estimation, and point map reconstruction.
Gheorghe Comanici, Eric Bieber, Mike Schaekermann +3300 authors
Gemini 2.X model family, including Gemini 2.5 Pro and Flash, offers superior coding, reasoning, and multimodal understanding capabilities across a range of computational efficiencies.
Anlin Zheng, Xin Wen, Xuanyang Zhang +5 authors
A novel image tokenizer built on pre-trained vision foundation models improves image reconstruction, generation quality, and token efficiency, enhancing autoregressive generation and class-conditional synthesis.
LG AI Research, Kyunghoon Bae, Eunbi Choi +38 authors
EXAONE 4.0 integrates non-reasoning and reasoning modes, supports agentic tool use, and offers multilingual capabilities, demonstrating superior performance in its class.
Yudong Jin, Sida Peng, Xuan Wang +6 authors
A sliding iterative denoising process enhances spatio-temporal consistency in 4D diffusion models for high-fidelity view synthesis from sparse-view videos.
Zhengqing Wang, Yuefan Wu, Jiacheng Chen +2 authors
A neural rendering approach using compressed light-field tokens (CLiFTs) achieves efficient rendering with adjustable token numbers, maintaining high quality and speed.
Youliang Zhang, Zhaoyang Li, Duomin Wang +6 authors
SpeakerVid-5M is a large-scale, high-quality dataset for audio-visual dyadic interactive virtual human generation, featuring diverse interaction types and data quality levels, along with an autoregressive video chat baseline and benchmark metrics.
Zhouqi Hua, Wenwei Zhang, Chengqi Lyu +4 authors
TAIL, a method that imitates Turing Machine execution processes, enhances the length generalization and performance of LLMs by synthesizing chain-of-thought data and reducing shortcut learning.
Ziang Cao, Zhaoxi Chen, Linag Pan +1 authors
PhysX addresses the gap in physical-grounded 3D asset generation by introducing PhysXNet, a physics-annotated dataset, and PhysXGen, a feed-forward framework that integrates physical knowledge into 3D generation.
Yiming Ren, Zhiqiang Lin, Yu Li +8 authors
The AnyCap Project introduces a framework, dataset, and evaluation protocol to enhance controllability and reliability in multimodal captioning.
Xinyi He, Qian Liu, Mingzhe Du +5 authors
SWE-Perf is a benchmark for evaluating LLMs in code performance optimization using real-world repository data, revealing significant gaps compared to expert performance.
Tiezheng Zhang, Yitong Li, Yu-cheng Chou +4 authors
The VLV auto-encoder framework uses pretrained vision and text models to create a cost-efficient, high-quality captioning system with reduced data requirements.
Mustafa Shukor, Louis Bethune, Dan Busbridge +4 authors
Scaling laws predict optimal data mixtures for large foundation models across different domains, improving performance and reducing trial-and-error.
Max Belitsky, Dawid J. Kopiczko, Michael Dorkenwald +3 authors
Cache steering improves language model reasoning through a single intervention in the key-value cache, enhancing both structure and performance without fine-tuning.
Mingxian Lin, Wei Huang, Yitang Li +6 authors
EmRACE-3K, a dataset of language-guided tasks in diverse environments, evaluates and improves embodied reasoning capabilities of vision-language models through zero-shot and fine-tuned performance.
Vladimir Bogachev, Vladimir Aletov, Alexander Molozhavenko +4 authors
RiemannLoRA addresses initialization and overparametrization in LoRA by treating LoRA matrices as a smooth manifold, improving convergence speed and performance in LLMs and diffusion models.
Alexander H. Liu, Andy Ehrenberg, Andy Lo +103 authors
Voxtral Mini and Voxtral Small are multimodal audio chat models that excel in understanding spoken audio and text, with a 32K context window for extended audio and conversation handling.
Yinsheng Li, Zhen Dong, Yi Shao
DrafterBench is an open-source benchmark for evaluating LLM agents in technical drawing revision, assessing their capabilities in structured data comprehension, function execution, instruction following, and critical reasoning.
Yulai Zhao, Haolin Liu, Dian Yu +3 authors
Generative reward models using large language models are vulnerable to superficial manipulations, leading to false positive rewards, and a new data augmentation strategy improves their robustness.
Hangjie Yuan, Weihua Chen, Jun Cen +11 authors
Lumos-1 is an autoregressive video generator that uses modified LLM architecture with MM-RoPE and AR-DF to handle spatiotemporal data efficiently, achieving competitive performance with reduced computational resources.
Orion Weller, Kathryn Ricci, Marc Marone +3 authors
The Ettin suite of models demonstrates that encoder-only and decoder-only architectures perform optimally in their respective tasks, with encoder-only models excelling at classification and retrieval, and decoder-only models at generation, and that adapting models to different tasks through continued training is less effective.
Yuncong Yang, Jiageng Liu, Zheyuan Zhang +5 authors
MindJourney enhances vision-language models with 3D reasoning by coupling them with a video diffusion-based world model, achieving improved performance on spatial reasoning tasks without fine-tuning.
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