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, Jinzheng He +11 authors
Qwen2.5-Omni is a multimodal model that processes text, images, audio, and video in a streaming fashion and generates text and speech using a dual-track architecture, achieving state-of-the-art performance on multimodal benchmarks.
50 篇论文 · 按点赞排序
Andrey Galichin, Alexey Dontsov, Polina Druzhinina +4 authors
A study uses Sparse Autoencoders to identify and enhance reasoning features in Large Language Models, providing insights into the mechanisms driving their reasoning capabilities.
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.
Junyu Luo, Weizhi Zhang, Ye Yuan +23 authors
A survey of large language model agents provides a structured taxonomy, exploring architectural principles, collaboration, evolution, evaluation, and applications, highlighting promising research directions.
Kaituo Feng, Kaixiong Gong, Bohao Li +5 authors
Video-R1, leveraging rule-based reinforcement learning and temporal information, enhances video reasoning in multimodal large language models using a combination of video and image data.
Yuchao Gu, Weijia Mao, Mike Zheng Shou
FAR, an autoregressive video model, achieves state-of-the-art results by balancing locality and long-range context with FlexRoPE and long short-term context modeling, addressing challenges in long-context video generation.
Eduard Allakhverdov, Elizaveta Goncharova, Andrey Kuznetsov
A method using autoencoder and Gumbel-Softmax selection identifies and retains only the most informative visual tokens, enabling efficient multimodal pruning with minimal performance loss.
WanTeam, Ang Wang, Baole Ai +59 authors
Wan, a comprehensive suite of video foundation models built on the diffusion transformer paradigm, advannces video generation by introducing a novel VAE, scalable pre-training strategies, and large-scale data curation, offering superior performance and versatility across various applications with both large and efficient models.
Zhengxi Lu, Yuxiang Chai, Yaxuan Guo +5 authors
Rule-based reinforcement learning enhances multimodal large language models for GUI action prediction, improving data efficiency and performance.
Gemma Team, Aishwarya Kamath, Johan Ferret +213 authors
Gemma 3 introduces vision capabilities, broader language coverage, and extended context length, featuring an optimized architecture and post-training enhancements to outperform previous versions.
Jian Zhang, Zhiyuan Wang, Zhangqi Wang +6 authors
A Multi-Agent framework with feedback mechanisms and Socratic guidance addresses multi-modal scientific problems and enhances reasoning and reflection, outperforming current models.
Zhi Hou, Tianyi Zhang, Yuwen Xiong +8 authors
Dita, a scalable framework using Transformer architectures and multimodal diffusion, achieves state-of-the-art performance with fine-grained action denoising and real-world adaptation.
Jiaheng Liu, Dawei Zhu, Zhiqi Bai +34 authors
This survey examines recent advancements in long-context language modeling, focusing on methods for obtaining, training, deploying, and evaluating models capable of processing extensive textual inputs.
Salaheddin Alzubi, Creston Brooks, Purva Chiniya +9 authors
Open Deep Search enhances open-source language models with a reasoning agent and search tool, achieving competitive performance compared to proprietary AI solutions.
Jian Zhang, Zhangqi Wang, Haiping Zhu +3 authors
A proposed Multi-Agent framework with Socratic guidance optimizes prompts for large language models by incorporating flexibility and efficient search mechanisms.
Baifeng Shi, Boyi Li, Han Cai +8 authors
PS3 enables high-resolution vision pre-training with reduced computational cost, improving multi-modal LLM performance and efficiency.
Jiwen Yu, Yiran Qin, Haoxuan Che +5 authors
A new framework for generative game engines using Interactive Generative Video to enable endless creation of realistic, interactive virtual environments with AI-driven capabilities.
Haoxiang Sun, Yingqian Min, Zhipeng Chen +5 authors
OlymMATH is a new benchmark for evaluating the complex reasoning capabilities of LLMs using challenging Olympiad-level problems in parallel English and Chinese.
John Joon Young Chung, Vishakh Padmakumar, Melissa Roemmele +2 authors
Post-training approaches incorporating deviation into direct preference optimization and odds ratio preference optimization enhance the output diversity of large language models without significantly compromising quality.
Kexian Tang, Junyao Gao, Yanhong Zeng +6 authors
LEGO-Puzzles evaluates the spatial understanding and sequential reasoning capabilities of Multimodal Large Language Models (MLLMs) through LEGO-based tasks, revealing significant limitations compared to human performance.
Dian Zheng, Ziqi Huang, Hongbo Liu +8 authors
VBench-2.0 introduces a benchmark for video generation that evaluates intrinsic faithfulness by assessing human fidelity, controllability, creativity, physics, and commonsense.
Jaihoon Kim, Taehoon Yoon, Jisung Hwang +1 authors
A new method enables efficient inference-time scaling for flow models through SDE-based generation, interpolant conversion, and Rollover Budget Forcing, improving sample quality and diversity.
Weihao Zeng, Yuzhen Huang, Qian Liu +4 authors
Zero reinforcement learning training across diverse base models improves reasoning accuracy and response length through strategic design adjustments, revealing novel cognitive behaviors in smaller models.
Gemini Robotics Team, Saminda Abeyruwan, Joshua Ainslie +115 authors
Gemini Robotics, an advanced Vision-Language-Action model based on Gemini 2.0, controls robots with multimodal reasoning, handling complex tasks and adapting to new environments and instructions.
Hongcheng Gao, Jiashu Qu, Jingyi Tang +6 authors
A comprehensive benchmark called HAVEN assesses hallucinations in video understanding tasks, and a video-thinking model using supervised reasoning fine-tuning and preference optimization reduces hallucinations and improves model accuracy.
Dvir Samuel, Matan Levy, Nir Darshan +2 authors
OmnimatteZero leverages pre-trained video diffusion models and zero-shot image inpainting to efficiently decompose videos into meaningful layers and perform object removal and recombination with real-time performance.
Zhicheng Lee, Shulin Cao, Jinxin Liu +5 authors
ReaRAG enhances large reasoning models by integrating retrieval capabilities with a novel data construction framework, improving factuality and robust reasoning for multi-hop question answering tasks.
Yitong Chen, Lingchen Meng, Wujian Peng +2 authors
CoMP, a multimodal pre-training pipeline with Continual Rotary Position Embedding and Alignment Loss, enhances visual representations for various tasks by aligning with language models.
Xiaoyu Tian, Sitong Zhao, Haotian Wang +5 authors
A multi-round thinking approach at test time improves the performance of large language models by iteratively refining their reasoning processes, enhancing accuracy on various benchmarks.
Yuqing Wang, Zhijie Lin, Yao Teng +4 authors
TokenBridge addresses the dilemma between discrete and continuous token representations in autoregressive visual generation by using post-training quantization to preserve visual details while maintaining modeling simplicity.
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