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。
Jin Xu, Zhifang Guo, Hangrui Hu +35 authors
Qwen3-Omni, a multimodal model, achieves state-of-the-art performance across text, image, audio, and video, using a Thinker-Talker MoE architecture and a lightweight causal ConvNet for efficient streaming synthesis.
50 篇论文 · 按点赞排序
Khalil Hennara, Muhammad Hreden, Mohamed Motasim Hamed +4 authors
Baseer, a vision-language model fine-tuned for Arabic document OCR, achieves state-of-the-art performance using a decoder-only strategy and a large-scale dataset, outperforming existing solutions with a WER of 0.25.
Jane Luo, Xin Zhang, Steven Liu +11 authors
A graph-driven framework called ZeroRepo uses the Repository Planning Graph (RPG) to generate complete software repositories from scratch, significantly outperforming existing baselines in terms of code size, functional coverage, and test pass rate.
Guochao Jiang, Wenfeng Feng, Guofeng Quan +4 authors
A curriculum reinforcement learning framework dynamically adjusts training sample difficulty based on reward variance, improving LLM performance on mathematical reasoning tasks.
Sicong Leng, Jing Wang, Jiaxi Li +12 authors
Variance-Aware Sampling and large-scale CoT data improve multimodal reasoning models by stabilizing RL fine-tuning and enhancing performance on benchmarks.
Yang Xiao, Mohan Jiang, Jie Sun +18 authors
LIMI demonstrates that sophisticated agentic intelligence can emerge from minimal, strategically curated demonstrations, outperforming data-intensive models on agency benchmarks.
Yizhou Wang, Chen Tang, Han Deng +29 authors
A scientific reasoning foundation model pre-trained on diverse scientific data supports multiple tasks and enhances cross-domain generalization and fidelity through specialized training techniques.
Thaddäus Wiedemer, Yuxuan Li, Paul Vicol +6 authors
Veo 3, a generative video model, exhibits zero-shot capabilities across various visual tasks, suggesting a trajectory towards becoming a unified, generalist vision foundation model.
Yuxiang Ji, Ziyu Ma, Yong Wang +3 authors
Tree-based Group Relative Policy Optimization (Tree-GRPO) enhances reinforcement learning for large language models by using tree search to improve rollouts and estimate grouped relative advantages, outperforming chain-based methods.
Team Seedream, Yunpeng Chen, Yu Gao +47 authors
Seedream 4.0 is a high-performance multimodal image generation system that integrates text-to-image synthesis, image editing, and multi-image composition using a diffusion transformer and VAE, achieving state-of-the-art results with efficient training and inference.
Siheng Li, Kejiao Li, Zenan Xu +33 authors
Reinforcement Learning on Pre-Training data (RLPT) optimizes large language models by autonomously exploring meaningful trajectories in pre-training data, improving generalizable reasoning skills without human annotation.
Jinshu Chen, Xinghui Li, Xu Bai +8 authors
OmniInsert addresses challenges in mask-free video insertion using a novel data pipeline, feature injection, progressive training, and context-aware rephrasing, outperforming commercial solutions.
Tianyu Yu, Zefan Wang, Chongyi Wang +31 authors
MiniCPM-V 4.5, a 8B parameter multimodal large language model, achieves high performance and efficiency through a unified 3D-Resampler architecture, a unified learning paradigm, and a hybrid reinforcement learning strategy.
Yanghao Li, Rui Qian, Bowen Pan +24 authors
Manzano is a unified multimodal LLM framework that integrates image and text processing using a hybrid tokenizer and diffusion decoder, achieving state-of-the-art performance in both understanding and generating visual content.
Henrique Schechter Vera, Sahil Dua, Biao Zhang +85 authors
EmbeddingGemma, a lightweight text embedding model based on Gemma 3, achieves state-of-the-art performance with fewer parameters through encoder-decoder initialization, geometric embedding distillation, and spread-out regularization.
Juntu Zhao, Wenbo Lu, Di Zhang +10 authors
A state-free policy using only visual observations achieves better spatial generalization and data efficiency in robot manipulation tasks compared to state-based policies.
Zinan Lin, Enshu Liu, Xuefei Ning +3 authors
Latent Zoning Network (LZN) unifies generative modeling, representation learning, and classification by creating a shared latent space for diverse data types.
Xilin Wei, Xiaoran Liu, Yuhang Zang +5 authors
SIM-CoT, a plug-and-play training module, introduces step-level supervision to stabilize and enrich the latent reasoning space of implicit Chain-of-Thought methods, enhancing their performance and efficiency.
Team Hunyuan3D, Bowen Zhang, Chunchao Guo +16 authors
Hunyuan3D-Omni is a unified 3D asset generation framework that accepts multiple conditioning signals, improving controllability and robustness in production workflows.
Ilya Alekseev, Roman Solomatin, Darina Rustamova +1 authors
AutoIntent is an automated machine learning tool for text classification that offers end-to-end automation, including embedding model selection, classifier optimization, and decision threshold tuning, and supports multi-label classification and out-of-scope detection.
Pierre Andrews, Amine Benhalloum, Gerard Moreno-Torres Bertran +21 authors
Meta Agents Research Environments (ARE) facilitate the creation and execution of complex environments for agent research, and Gaia2, a benchmark built on ARE, evaluates general agent capabilities in dynamic, asynchronous settings.
Weihan Peng, Yuling Shi, Yuhang Wang +3 authors
SWE-QA is a repository-level code question answering benchmark that addresses the complexity of real-world software repositories, evaluated using advanced LLMs and an agentic framework.
Sunhao Dai, Jiakai Tang, Jiahua Wu +13 authors
OnePiece integrates LLM-style context engineering and reasoning into industrial search and recommendation systems, achieving significant improvements in key business metrics.
Sherwin Bahmani, Tianchang Shen, Jiawei Ren +10 authors
A self-distillation framework converts implicit 3D knowledge from video diffusion models into an explicit 3D Gaussian Splatting representation, enabling 3D scene generation from text or images.
Songsong Yu, Yuxin Chen, Hao Ju +15 authors
Research investigates visual spatial reasoning in vision-language models, highlighting gaps between perceptual and reasoning capabilities, and introduces SIBench as a benchmark for future research.
Yidong Wang, Yunze Song, Tingyuan Zhu +11 authors
TrustJudge, a probabilistic framework, addresses inconsistencies in LLM-as-a-judge evaluation by using distribution-sensitive scoring and likelihood-aware aggregation, improving accuracy and reliability.
Chuan Fang, Heng Li, Yixun Liang +6 authors
SpatialGen, a multi-view multi-modal diffusion model, generates realistic and semantically consistent 3D indoor scenes using a large synthetic dataset, outperforming previous methods.
Wenke Huang, Quan Zhang, Yiyang Fang +11 authors
Mixed Advantage Policy Optimization (MAPO) dynamically reweights the advantage function to improve trajectory ranking in reinforcement learning for foundation models.
Yunheng Li, Jing Cheng, Shaoyong Jia +4 authors
TempSamp-R1, a reinforcement fine-tuning framework, enhances multimodal large language models for video temporal grounding by using off-policy supervision and a hybrid Chain-of-Thought training paradigm, achieving state-of-the-art performance on benchmark datasets.
Weijie Wang, Yeqing Chen, Zeyu Zhang +7 authors
VolSplat, a voxel-aligned Gaussian prediction method, improves novel view synthesis by overcoming pixel alignment limitations and enhancing 3D reconstruction quality.
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