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
Hyunmin Cho, Donghoon Ahn, Susung Hong +3 authors
Tangential Amplifying Guidance (TAG) improves diffusion model sample quality by directly amplifying tangential components of estimated scores without modifying the model architecture.
Caorui Li, Yu Chen, Yiyan Ji +39 authors
OmniVideoBench is a comprehensive benchmark for evaluating audio-visual reasoning in multimodal large language models, addressing modality complementarity and logical consistency.
Yu Qi, Haibo Zhao, Ziyu Guo +17 authors
BEAR is a comprehensive benchmark evaluating multimodal large language models' embodied capabilities, and BEAR-Agent enhances these models by integrating pretrained vision models, improving performance across various tasks.
Quan Nguyen-Tri, Mukul Ranjan, Zhiqiang Shen
Elastic-Cache optimizes key-value cache management in diffusion large language models to reduce decoding latency without sacrificing prediction accuracy.
Wenkai Yang, Weijie Liu, Ruobing Xie +4 authors
LaSeR, a reinforcement learning algorithm, enhances Large Language Models by aligning last-token self-rewarding scores with verifier-based reasoning rewards, improving reasoning performance and inference-time scaling.
Junhao Zhuang, Shi Guo, Xin Cai +4 authors
Diffusion models have recently advanced video restoration, but applying them to real-world video super-resolution (VSR) remains challenging due to high latency, prohibitive computation, and poor generalization to ultra-high resolutions. Our goal in this work is to make diffusion-based VSR practical by achieving efficiency, scalability, and real-time performance. To this end, we propose FlashVSR, the first diffusion-based one-step streaming framework towards real-time VSR. FlashVSR runs at approximately 17 FPS for 768x1408 videos on a single A100 GPU by combining three complementary innovations: (i) a train-friendly three-stage distillation pipeline that enables streaming super-resolution, (ii) locality-constrained sparse attention that cuts redundant computation while bridging the train-test resolution gap, and (iii) a tiny conditional decoder that accelerates reconstruction without sacrificing quality. To support large-scale training, we also construct VSR-120K, a new dataset with 120k videos and 180k images. Extensive experiments show that FlashVSR scales reliably to ultra-high resolutions and achieves state-of-the-art performance with up to 12x speedup over prior one-step diffusion VSR models. We will release the code, pretrained models, and dataset to foster future research in efficient diffusion-based VSR.
Terry Yue Zhuo, Xiaolong Jin, Hange Liu +37 authors
BigCodeArena is an open human evaluation platform for code generation that enables real-time execution and interaction, revealing preferences and capabilities of LLMs in coding tasks.
Sihui Ji, Xi Chen, Xin Tao +2 authors
PhysMaster enhances video generation by integrating physical knowledge through PhysEncoder, using reinforcement learning and Direct Preference Optimization to improve physics-awareness.
Haomin Wang, Jinhui Yin, Qi Wei +12 authors
A unified multimodal large language model (MLLM) for SVG understanding, editing, and generation leverages a comprehensive dataset and benchmark to achieve superior performance across various tasks.
Jinghao Zhang, Naishan Zheng, Ruilin Li +4 authors
RLFR uses flow rewards derived from latent space to improve reinforcement learning with verifiable rewards, demonstrating reliable reward shaping and efficient context comprehension.
Yinxi Li, Yuntian Deng, Pengyu Nie
Misaligned tokenization in large language models for code leads to inconsistent model behavior, necessitating grammar-aware tokenization.
Guoqing Wang, Sunhao Dai, Guangze Ye +5 authors
Information Gain-based Policy Optimization (IGPO) enhances multi-turn reasoning in large language models by providing dense intrinsic rewards derived from the model's belief updates, improving accuracy and sample efficiency.
Devvrit Khatri, Lovish Madaan, Rishabh Tiwari +6 authors
A systematic study defines a framework for analyzing and predicting reinforcement learning scaling in large language models, identifying key design choices that affect compute efficiency and proposing a best-practice recipe.
Zhaochen Yu, Ling Yang, Jiaru Zou +2 authors
Agentic reinforcement learning enhances LLMs' reasoning ability through real datasets, exploration techniques, and a deliberative strategy, achieving strong performance with smaller models.
Wenwen Tong, Hewei Guo, Dongchuan Ran +23 authors
InteractiveOmni is a unified omni-modal large language model for audio-visual multi-turn interactions, offering comprehensive understanding and speech generation capabilities with efficient parameter usage.
Zhepeng Cen, Haolin Chen, Shiyu Wang +7 authors
A scalable data engine converts large-scale pre-training documents into diverse question-answer pairs for reinforcement learning, significantly improving model performance and efficiency.
Ahmed Heakl, Martin Gubri, Salman Khan +2 authors
Large Language Models (LLMs) process every token through all layers of a transformer stack, causing wasted computation on simple queries and insufficient flexibility for harder ones that need deeper reasoning. Adaptive-depth methods can improve efficiency, but prior approaches rely on costly inference-time search, architectural changes, or large-scale retraining, and in practice often degrade accuracy despite efficiency gains. We introduce Dr.LLM, Dynamic routing of Layers for LLMs, a retrofittable framework that equips pretrained models with lightweight per-layer routers deciding to skip, execute, or repeat a block. Routers are trained with explicit supervision: using Monte Carlo Tree Search (MCTS), we derive high-quality layer configurations that preserve or improve accuracy under a compute budget. Our design, windowed pooling for stable routing, focal loss with class balancing, and bottleneck MLP routers, ensures robustness under class imbalance and long sequences. On ARC (logic) and DART (math), Dr.LLM improves accuracy by up to +3.4%p while saving 5 layers per example on average. Routers generalize to out-of-domain tasks (MMLU, GSM8k, AIME, TruthfulQA, SQuADv2, GPQA, PIQA, AGIEval) with only 0.85% accuracy drop while retaining efficiency, and outperform prior routing methods by up to +7.7%p. Overall, Dr.LLM shows that explicitly supervised routers retrofit frozen LLMs for budget-aware, accuracy-driven inference without altering base weights.
Xinhang Liu, Yuxi Xiao, Donny Y. Chen +4 authors
Trace Anything, a neural network, predicts video trajectories in a single pass, achieving state-of-the-art performance and demonstrating efficiency and emergent abilities like motion forecasting.
Haoran Feng, Dizhe Zhang, Xiangtai Li +2 authors
DiT360 framework enhances panoramic image generation by hybrid training on perspective and panoramic data, incorporating cross-domain knowledge and hybrid supervision to improve boundary consistency and image fidelity.
Youngrok Park, Hojung Jung, Sangmin Bae +1 authors
Diffusion models have achieved remarkable success as generative models. However, even a well-trained model can accumulate errors throughout the generation process. These errors become particularly problematic when arbitrary guidance is applied to steer samples toward desired properties, which often breaks sample fidelity. In this paper, we propose a general solution to address the off-manifold phenomenon observed in diffusion models. Our approach leverages a time predictor to estimate deviations from the desired data manifold at each timestep, identifying that a larger time gap is associated with reduced generation quality. We then design a novel guidance mechanism, `Temporal Alignment Guidance' (TAG), attracting the samples back to the desired manifold at every timestep during generation. Through extensive experiments, we demonstrate that TAG consistently produces samples closely aligned with the desired manifold at each timestep, leading to significant improvements in generation quality across various downstream tasks.
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