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。
MiniMax, Aonian Li, Bangwei Gong +87 authors
The MiniMax-01 series, including MiniMax-Text-01 and MiniMax-VL-01, offer superior long-context processing and match state-of-the-art model performance with longer context windows through lightning attention, Mixture of Experts (MoE), and efficient parallel strategies.
50 篇论文 · 按点赞排序
Zhenru Zhang, Chujie Zheng, Yangzhen Wu +6 authors
Process Reward Models improve performance in mathematical reasoning by integrating Monte Carlo estimation with LLM-as-a-judge, addressing biases in Best-of-N evaluation and step-wise error identification.
Yifan Zhang, Yifeng Liu, Huizhuo Yuan +4 authors
T6, a new Transformer architecture using Tensor Product Attention, improves language model performance and memory efficiency, allowing for longer sequence processing.
Nanye Ma, Shangyuan Tong, Haolin Jia +8 authors
Inference-time scaling of diffusion models improves sample quality beyond increasing denoising steps through better noise selection and verification.
Soyeong Jeong, Kangsan Kim, Jinheon Baek +1 authors
VideoRAG dynamically retrieves relevant videos based on queries and utilizes both visual and textual information for output generation, outperforming existing methods in response accuracy.
Omkar Thawakar, Dinura Dissanayake, Ketan More +12 authors
A framework for evaluating and improving step-by-step visual reasoning in large language models using a specialized benchmark and a novel multimodal model trained with curriculum learning.
Zhengyang Tang, Ziniu Li, Zhenyang Xiao +8 authors
SCRIT, a self-evolving critique framework, enhances LLMs' critique capabilities using synthetic data and self-validation, achieving significant improvements in critique-correction and error identification benchmarks.
Zhiheng Liu, Ka Leong Cheng, Xi Chen +7 authors
MangaNinjia, a diffusion model-based approach, enhances line art colorization with patch shuffling and point-driven control for precise detail and color matching.
Stefan Baack, Stella Biderman, Kasia Odrozek +36 authors
Many AI companies are training their large language models (LLMs) on data without the permission of the copyright owners. The permissibility of doing so varies by jurisdiction: in countries like the EU and Japan, this is allowed under certain restrictions, while in the United States, the legal landscape is more ambiguous. Regardless of the legal status, concerns from creative producers have led to several high-profile copyright lawsuits, and the threat of litigation is commonly cited as a reason for the recent trend towards minimizing the information shared about training datasets by both corporate and public interest actors. This trend in limiting data information causes harm by hindering transparency, accountability, and innovation in the broader ecosystem by denying researchers, auditors, and impacted individuals access to the information needed to understand AI models. While this could be mitigated by training language models on open access and public domain data, at the time of writing, there are no such models (trained at a meaningful scale) due to the substantial technical and sociological challenges in assembling the necessary corpus. These challenges include incomplete and unreliable metadata, the cost and complexity of digitizing physical records, and the diverse set of legal and technical skills required to ensure relevance and responsibility in a quickly changing landscape. Building towards a future where AI systems can be trained on openly licensed data that is responsibly curated and governed requires collaboration across legal, technical, and policy domains, along with investments in metadata standards, digitization, and fostering a culture of openness.
Mingjie Pan, Jiyao Zhang, Tianshu Wu +3 authors
A novel object-centric representation bridges the gap between Vision-Language Models' high-level reasoning and the low-level precision needed for robotic manipulation tasks, enabling zero-shot generalization across diverse tasks.
Qi Sun, Edoardo Cetin, Yujin Tang
A self-adaptive framework for large language models uses reinforcement learning to dynamically adjust task-specific components during inference, enhancing adaptability and performance with efficiency.
Alejandro Lozano, Min Woo Sun, James Burgess +13 authors
BIOMEDICA is a scalable framework that creates a large, publicly accessible dataset of image-text pairs from PubMed Central, enabling state-of-the-art performance across various biomedical tasks with efficient CLIP-style models.
Qian Chen, Yafeng Chen, Yanni Chen +33 authors
MinMo, a multimodal large language model, integrates speech and text processing to achieve state-of-the-art performance in voice comprehension and generation, while enabling full-duplex conversation and instruction-following capabilities.
Zekun Xi, Wenbiao Yin, Jizhan Fang +7 authors
OmniThink is a machine writing framework that simulates human iterative learning to enhance the quality and depth of generated long-form articles.
Yifei Li, Junbo Niu, Ziyang Miao +12 authors
OVO-Bench is a new benchmark for evaluating the temporal awareness of online video language models across three scenarios: backward tracing, real-time understanding, and forward active responding.
Fengli Xu, Qianyue Hao, Zefang Zong +17 authors
Large Language Models are being enhanced for complex reasoning through thought sequences, reinforcement learning, and scaling techniques, leading to advanced reasoning capabilities exemplified by OpenAI's o1 series.
Dewei Zhou, Ji Xie, Zongxin Yang +1 authors
3DIS-FLUX extends the 3DIS framework with FLUX for improved text-to-image generation, achieving better performance and image quality through depth map control and attention mask manipulation.
Shanchuan Lin, Xin Xia, Yuxi Ren +3 authors
Adversarial Post-Training improves the quality and speed of one-step video and image generation using diffusion models.
Philippe Hansen-Estruch, David Yan, Ching-Yao Chung +7 authors
Replacing the convolutional backbone with a Vision Transformer for tokenization yields competitive auto-encoder performance and sets benchmarks for class-conditional video generation when integrated with Diffusion Transformers.
Miran Heo, Min-Hung Chen, De-An Huang +5 authors
Omni-RGPT uses Token Mark with region prompts to achieve consistent and robust region-level multimodal understanding, supporting tasks like reasoning, captioning, and referring expression comprehension.
Saman Motamed, Laura Culp, Kevin Swersky +2 authors
A new benchmark dataset, Physics-IQ, shows that current AI video models achieve visual realism without understanding underlying physical principles, highlighting the need for further advancements.
Zhongzhen Huang, Gui Geng, Shengyi Hua +5 authors
Inference-time scaling in large language models enhances medical reasoning performance by increasing the length of reasoning chains and adhering to the hypothetico-deductive method in generating differential diagnoses.
Michael Toker, Ido Galil, Hadas Orgad +4 authors
An analysis of padding tokens' influence in T2I diffusion models identifies scenarios where they impact image generation and relationships with architecture and training process.
Junfei Xiao, Feng Cheng, Lu Qi +5 authors
A large-scale cooking video dataset and a Long Narrative Video Director model improve visual and semantic coherence in video generation by aligning text and image embeddings.
Kuicai Dong, Yujing Chang, Xin Deik Goh +3 authors
A new benchmark, MMDocIR, is introduced for evaluating multi-modal document retrieval systems, highlighting the superior performance of visual retrievers over text-only methods and emphasizing the importance of visual elements.
You Li, Heyu Huang, Chi Chen +8 authors
Migician, a multi-image grounding model, outperforms existing MLLMs in complex multi-image scenarios using the MGrounding-630k dataset and MIG-Bench benchmark.
Karl Pertsch, Kyle Stachowicz, Brian Ichter +6 authors
A new tokenization method called FAST, based on the discrete cosine transform, enables training of autoregressive vision-language-action policies for high-frequency robotic tasks with improved performance and reduced training time.
Yin Fang, Xinle Deng, Kangwei Liu +5 authors
InstructCell, a multi-modal AI copilot, enables researchers to perform single-cell analysis using natural language instructions, improving accessibility and performance compared to existing tools.
Jialong Wu, Wenbiao Yin, Yong Jiang +6 authors
WebWalkerQA assesses LLMs' ability to traverse websites for high-quality data, showing enhancements when combined with RAG using the WebWalker multi-agent framework.
Haozhe Xie, Zhaoxi Chen, Fangzhou Hong +1 authors
CityDreamer4D generates realistic 4D cities by separating dynamic and static elements using customized neural fields and compositional generative models.
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