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。
Zhiyu Li, Shichao Song, Chenyang Xi +36 authors
MemOS, a memory operating system for Large Language Models, addresses memory management challenges by unifying plaintext, activation-based, and parameter-level memories, enabling efficient storage, retrieval, and continual learning.
50 篇论文 · 按点赞排序
Yukang Chen, Wei Huang, Baifeng Shi +11 authors
A framework scales vision-language models for long video reasoning using reinforcement learning, achieving strong performance on benchmarks and demonstrating consistent gains with increased video frames.
Vera Soboleva, Aibek Alanov, Andrey Kuznetsov +1 authors
T-LoRA, a timestep-dependent low-rank adaptation framework, enhances diffusion model personalization with a dynamic fine-tuning strategy and orthogonal initialization, improving concept fidelity and text alignment in data-limited settings.
David Bensaïd, Noam Rotstein, Roy Velich +2 authors
SingLoRA, a reformulated low-rank adaptation method, enhances parameter-efficient fine-tuning by learning a single low-rank matrix and its transpose, ensuring stable optimization and reducing parameter count.
Yushen Zuo, Qi Zheng, Mingyang Wu +10 authors
4KAgent, a unified agentic super-resolution system, enhances low-resolution images to 4K using a profiling module, perception agent, and restoration agent, achieving state-of-the-art performance across various imaging domains.
Rui-Jie Zhu, Tianhao Peng, Tianhao Cheng +30 authors
Latent reasoning in Large Language Models (LLMs) performs multi-step inference in continuous hidden states, enhancing reasoning capabilities without token-level supervision, and includes methodologies like activation-based recurrence and infinite-depth reasoning via masked diffusion models.
Hippolyte Gisserot-Boukhlef, Nicolas Boizard, Manuel Faysse +5 authors
A biphasic training strategy combining Causal Language Modeling and Masked Language Modeling yields optimal text representation performance, especially when initialized with pretrained CLM models.
Yu Wang, Xi Chen
MIRIX, a modular multi-agent memory system, enhances language models' memory capabilities by integrating diverse memory types and a dynamic framework, achieving superior performance in multimodal and long-form conversation benchmarks.
Wei Shen, Jiangbo Pei, Yi Peng +7 authors
Skywork-R1V3, an open-source vision-language model, enhances visual reasoning through a post-training reinforcement learning framework, achieving state-of-the-art performance on multimodal reasoning tasks.
Yunhan Yang, Yufan Zhou, Yuan-Chen Guo +7 authors
OmniPart generates part-aware 3D objects with high semantic decoupling and structural cohesion using an autoregressive structure planning module and a spatially-conditioned rectified flow model.
Ke Fan, Shunlin Lu, Minyue Dai +6 authors
A new dataset and evaluation framework for text-to-motion generation achieve zero-shot generalization using a large-scale model.
Ziyang Miao, Qiyu Sun, Jingyuan Wang +4 authors
A framework called Easy Dataset synthesizes fine-tuning data from unstructured documents using a GUI and LLMs, improving domain-specific performance of LLMs while maintaining general knowledge.
Dheeraj Vattikonda, Santhoshi Ravichandran, Emiliano Penaloza +13 authors
A study on compute allocation for post-training LLM-based web agents finds that combining supervised fine-tuning with on-policy reinforcement learning improves performance and reduces computational costs compared to using either method alone.
Haochen Wang, Xiangtai Li, Zilong Huang +9 authors
TreeBench evaluates visual grounded reasoning with a focus on subtle targets, traceable evidence, and second-order reasoning, while TreeVGR enhances this reasoning using reinforcement learning.
Zhenhailong Wang, Xuehang Guo, Sofia Stoica +8 authors
PAPO, an extension of GRPO, enhances multimodal reasoning by integrating perception-aware supervision, leading to improved performance and reduced perception errors in tasks with high vision dependency.
Jeongseok Hyun, Sukjun Hwang, Su Ho Han +6 authors
A training-free method merges spatio-temporal tokens in video large language models to reduce computational cost while maintaining accuracy.
Meng Wei, Chenyang Wan, Xiqian Yu +9 authors
StreamVLN, a streaming VLN framework, uses a hybrid slow-fast context modeling strategy to achieve state-of-the-art performance with low latency and efficient resource usage.
Zhongyuan Peng, Yifan Yao, Kaijing Ma +16 authors
CriticLean, a critic-guided reinforcement learning framework, enhances the evaluation of formalizations in automated theorem proving by actively learning and distinguishing semantically correct formalizations.
Wenyao Zhang, Hongsi Liu, Zekun Qi +10 authors
DreamVLA integrates comprehensive world knowledge forecasting with a block-wise structured attention mechanism and diffusion-based transformer to improve action prediction and generalization in robot manipulation tasks.
JingLi Lin, Chenming Zhu, Runsen Xu +4 authors
OST-Bench evaluates multimodal large language models on online spatio-temporal reasoning tasks, highlighting challenges in dynamic spatial understanding and long-term memory retrieval.
Yutian Chen, Shi Guo, Tianshuo Yang +4 authors
A novel capturing and processing system using low FPS cameras and a video-diffusion-based generative model enhances high-speed 4D reconstruction.
Wanhua Li, Yujie Zhao, Minghan Qin +4 authors
LangSplatV2 improves the speed and accuracy of 3D open-vocabulary text querying by eliminating the decoder and using sparse Gaussian splatting.
Shihan Dou, Shichun Liu, Yuming Yang +19 authors
A novel reward modeling approach, Policy Discriminative Learning (POLAR), enhances reward model performance and generalization in reinforcement learning by focusing on relative policy differences.
Ziyue Li, Yang Li, Tianyi Zhou
A method using chain-of-layers (CoLa) and Monte Carlo Tree Search (MCTS) optimizes the architecture of a pretrained large language model for individual samples, improving inference efficiency and performance.
Rahul Ramachandran, Ali Garjani, Roman Bachmann +3 authors
Popular multimodal foundation models perform respectably across various computer vision tasks when adapted through prompt chaining, though they fall short of specialist models and exhibit quirks in image generation.
Haoyu Wu, Diankun Wu, Tianyu He +4 authors
Geometry Forcing enhances video diffusion models by aligning their intermediate representations with geometric features from a pretrained model, improving visual quality and 3D consistency.
Shitian Zhao, Haoquan Zhang, Shaoheng Lin +4 authors
PyVision enables MLLMs to autonomously generate, execute, and refine Python-based tools for visual reasoning, achieving significant performance improvements across benchmarks.
Akio Kodaira, Tingbo Hou, Ji Hou +2 authors
StreamDiT, a streaming video generation model using transformer-based diffusion with flow matching and adaLN DiT, achieves real-time performance at 16 FPS with 4B parameters and multistep distillation.
Peisong Wang, Ruotian Ma, Bang Zhang +13 authors
RLVER, an end-to-end reinforcement learning framework using verifiable emotion rewards, enhances emotional intelligence in LLMs without sacrificing cognitive abilities.
Shangzhan Li, Zefan Wang, Ye He +8 authors
AutoTriton, a reinforcement learning model, automates the generation of high-performance kernels for GPU programming by combining supervised fine-tuning and Group Relative Policy Optimization.
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