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。
Zesen Cheng, Hang Zhang, Kehan Li +6 authors
A tile-based computation strategy optimizes contrastive loss by partitioning calculations, enabling large batch sizes without significant memory or speed sacrifices.
50 篇论文 · 按点赞排序
Chin-Yang Lin, Chung-Ho Wu, Chang-Han Yeh +3 authors
FrugalNeRF addresses few-shot NeRF challenges by using weight-sharing voxels and a cross-scale adaptation scheme, improving quality and reducing training time.
Shuangrui Ding, Rui Qian, Xiaoyi Dong +6 authors
SAM2Long enhances SAM 2's video segmentation by using a memory module and constrained tree search to improve accuracy in complex, long-term videos.
Maosong Cao, Alexander Lam, Haodong Duan +3 authors
CompassJudger-1 is an open-source, all-in-one judge LLM that evaluates LLMs through unitary scoring, model comparisons, critiques, and various tasks, supported by the JudgerBench benchmark.
Yuzhe Yang, Yifei Zhang, Yan Hu +10 authors
A user-centric framework evaluates LLMs in financial tasks through human feedback and task-specific interactions, achieving high correlation with human preferences.
Abhishek Thakur
AutoTrain, an open-source tool, streamlines training and fine-tuning of various machine learning models across multiple tasks and modalities using best practices.
Rongyao Fang, Chengqi Duan, Kun Wang +7 authors
PUMA, a unified multimodal large language model, addresses varying granularity demands in visual content generation tasks through the integration of multi-granular visual features.
Baixiang Huang, Canyu Chen, Xiongxiao Xu +2 authors
HalluEditBench benchmarks knowledge editing methods in correcting hallucinations in Large Language Models (LLMs) using a comprehensive dataset and evaluation framework.
Mingan Lin, Fan Yang, Yanjun Shen +22 authors
Baichuan Alignment provides comprehensive insights into alignment methodologies used in Baichuan models, detailing improvements through Prompt Augmentation System, Supervised Fine-Tuning, and Preference Alignment across various benchmarks.
Junyu Luo, Xiao Luo, Xiusi Chen +3 authors
A semi-supervised fine-tuning framework named SemiEvol enhances LLM adaptation using both labeled and unlabeled data, showing improved performance through bi-level knowledge propagation and collaborative learning.
Long Xing, Qidong Huang, Xiaoyi Dong +8 authors
PyramidDrop, a novel strategy for reducing visual token redundancy in large vision-language models, enhances training and inference efficiency with minimal performance loss.
Xiang Yue, Yueqi Song, Akari Asai +7 authors
Pangea, a multilingual multimodal LLM, is trained on a diverse 6M instruction dataset across 39 languages and evaluated using PangeaBench, demonstrating superior performance in multilingual and cross-cultural settings.
Cheng-De Fan, Chen-Wei Chang, Yi-Ruei Liu +4 authors
SpectroMotion improves upon 3D Gaussian Splatting by integrating physical-based rendering and deformation fields, incorporating a residual correction technique and deformable environment map to accurately reconstruct and synthesize dynamic specular scenes.
Hyungjoo Chae, Namyoung Kim, Kai Tzu-iunn Ong +6 authors
A World-model-augmented web agent improves decision-making by simulating action outcomes, overcoming training challenges with free-form natural language descriptions of state differences.
Zecheng Tang, Zechen Sun, Juntao Li +2 authors
LOGO, a preference optimization strategy, enhances LCM generation performance and context window size by efficiently training with limited data.
Yuyang Ding, Xinyu Shi, Xiaobo Liang +3 authors
ScaleQuest is a scalable data synthesis method that generates high-quality mathematical reasoning datasets, improving the performance of open-source models.
Wen Wang, Qiuyu Wang, Kecheng Zheng +6 authors
Framer is an interactive frame interpolation system that allows users to customize transitions between images using keypad manipulation and includes an autopilot mode for automatic keypoint estimation.
Baiqi Li, Zhiqiu Lin, Wenxuan Peng +7 authors
VLMs perform poorly on a new benchmark, NaturalBench, which uses natural images and questions, highlighting challenges in understanding attribute bindings, object relationships, and advanced reasoning.
Jialu Li, Yuanzhen Li, Neal Wadhwa +5 authors
A generative AI-driven video game, Unbounded, offers an open-ended character life simulation using advanced generative models and image generation techniques.
Donghao Zhou, Jiancheng Huang, Jinbin Bai +5 authors
MagicTailor addresses challenges in T2I models by dynamically targeting and balancing specific visual components through DM-Deg and DS-Bal, enhancing precise control over image generation.
Ziyu Liu, Yuhang Zang, Xiaoyi Dong +7 authors
MIA-DPO is a method for visual preference alignment that improves handling of multi-image inputs by extending single-image data and using attention values to refine chosen/rejected pairs without extra annotations or data.
Yizhao Gao, Zhichen Zeng, Dayou Du +5 authors
SeerAttention introduces a learnable attention mechanism that adaptsively selects significant blocks in attention maps, balancing accuracy and speed, and achieves superior performance over static or heuristic-based sparse attention methods.
Yantao Liu, Zijun Yao, Rui Min +3 authors
RM-Bench evaluates reward models based on sensitivity to subtle content differences and style biases, providing a reliable metric for selecting reward models in RLHF and Inference Scaling Laws.
Xiaoqian Shen, Yunyang Xiong, Changsheng Zhao +14 authors
LongVU is a spatiotemporal adaptive compression mechanism that enhances video understanding by reducing token redundancy while preserving visual details, enabling long video processing within the context size constraints of large language models.
Ruohong Zhang, Bowen Zhang, Yanghao Li +6 authors
Enhancing vision language models with detailed rationales through distillation and reinforcement learning improves their chain-of-thought reasoning and generalization.
Hao Xiang, Bowen Yu, Hongyu Lin +6 authors
Self-Steering Optimization autonomously generates accurate preference signals without manual annotation, improving policy and reward model performance.
Jihao Zhao, Zhiyuan Ji, Pengnian Qi +4 authors
Meta-Chunking improves RAG's performance in text chunking for knowledge-intensive tasks by using LLMs to dynamically segment text, balancing fine-grained and coarse-grained chunking.
ZiDong Wang, Zeyu Lu, Di Huang +3 authors
Flexible Vision Transformers (FiT and FiTv2) enable unrestricted resolution and aspect ratio image generation through innovative modular design, improving flexibility and performance across various resolutions.
Xinyou Wang, Zaixiang Zheng, Fei Ye +3 authors
DPLM-2 is a multimodal protein foundation model that simultaneously generates amino acid sequences and their 3D structures using a discrete diffusion approach, improving performance in tasks like folding and scaffold prediction.
Chris Yuhao Liu, Liang Zeng, Jiacai Liu +6 authors
Curated preference datasets and selection strategies enhanced reward models, leading to improved performance on the RewardBench leaderboard.
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