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
Youngjoon Yu, Sangyun Chung, Byung-Kwan Lee +1 authors
A benchmark called SPARK is established to evaluate multi-vision sensory perception and reasoning in large-scale vision-language models, revealing deficiencies in understanding physical sensor characteristics.
Valentinos Pariza, Mohammadreza Salehi, Gertjan Burghouts +2 authors
NeCo, a novel self-supervised learning method using patch neighbor consistency, improves pretrained representations with minimal computational cost and achieves state-of-the-art results in semantic segmentation.
Jian Chen, Vashisth Tiwari, Ranajoy Sadhukhan +4 authors
MagicDec improves high-throughput inference for long-context LLMs by optimizing speculative decoding and using draft models with sparse KV cache.
Jonathan Light, Min Cai, Weiqin Chen +5 authors
Strategist uses LLMs and self-play simulations to enhance agents' strategic skills in multi-agent games, outperforming traditional RL and other LLM-based methods.
Chao Xu, Ang Li, Linghao Chen +4 authors
A novel method, SpaRP, reconstructs 3D textured meshes and estimates camera poses from sparse-view images by leveraging and finetuning 2D diffusion models.
Zhaochen Su, Jun Zhang, Xiaoye Qu +6 authors
ConflictBank is a benchmark for evaluating knowledge conflicts in LLMs by assessing retrieved knowledge, model-encoded knowledge, and their interplay across different model families and causes like misinformation and temporal discrepancies.
Yuanyang Yin, Yaqi Zhao, Yajie Zhang +7 authors
Supervised Embedding Alignment (SEA) improves Multimodal Large Language Models by aligning visual tokens with the LLM's embedding space, enhancing performance and interpretability.
Tej Deep Pala, Vernon Y. H. Toh, Rishabh Bhardwaj +1 authors
Ferret is a novel approach that enhances the efficiency and effectiveness of adversarial prompt generation for large language models by using multiple mutations and a reward model scoring function.
Haoning Wu, Shaocheng Shen, Qiang Hu +3 authors
MegaFusion enhances diffusion-based text-to-image generation for high-resolution images using a truncate and relay strategy with dilated convolutions and noise re-scheduling.
Baixiang Huang, Canyu Chen, Kai Shu
A literature review examines challenges and methods in authorship attribution for human and machine-generated texts, including detection and co-authored content, emphasizing generalization and explainability.
Yushi Yang, Andrew M. Bean, Robert McCraith +1 authors
Evaluating curriculum-based and non-curriculum-based learning strategies across multiple large language models for medical question answering shows variable effectiveness, with model-generated question difficulty measures potentially outperforming human-defined ones.
Ruofan Liang, Zan Gojcic, Merlin Nimier-David +4 authors
A personalized diffusion model guides a physically based inverse rendering process to enhance the photorealistic insertion of virtual objects into real-world scenes.
Yunxin Li, Haoyuan Shi, Baotian Hu +5 authors
Anim-Director, an autonomous animation-making agent, uses large multimodal models to create consistent and coherent animations from concise narratives.
Jonathan Roberts, Kai Han, Samuel Albanie
GRAB is a synthetic graph analysis benchmark designed to challenge large multimodal models (LMMs) across various tasks and graph properties, with the top-performing model achieving only 21.7% accuracy.
Niyar R Barman, Krish Sharma, Ashhar Aziz +7 authors
Visual paraphrase attacks using state-of-the-art captioning and diffusion systems can effectively remove watermarks from AI-generated images, highlighting the vulnerability of existing watermarking techniques.
Chenhan Yuan, Fei Huang, Ru Peng +4 authors
Non-disruptive parameter insertion into transformer architecture enhances large language models by predicting calibration signals without additional overhead.
Dennis Fedorishin, Lie Lu, Srirangaraj Setlur +1 authors
A technique for automatically identifying and generating audio match cuts in videos is explored, with a focus on developing a self-supervised audio representation and evaluating audio blending methods.
Mingrui Wu, Oucheng Huang, Jiayi Ji +6 authors
TraDiffusion, a training-free trajectory-based approach, allows precise control over image generation using mouse trajectories by adjusting latent variables with a custom energy function.
Junwon Lee, Jaekwon Im, Dabin Kim +1 authors
Video-Foley achieves top performance in video-to-sound generation by using RMS for temporal control and semantic timbre prompts without human annotation.
Thorsten Hellert, João Montenegro, Andrea Pollastro
PhysBERT is a physics-specific text embedding model pre-trained on arXiv physics papers, surpassing general-purpose models in physics-related tasks and fine-tuning for subdomains.
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