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。
598 篇论文 · 按点赞排序
Fanqing Meng, Jiaqi Liao, Xinyu Tan +7 authors
PhyGenBench and PhyGenEval assess text-to-video models' understanding of physical commonsense, revealing shortcomings that cannot be fully addressed by scaling models or prompt engineering.
Xingyi Yang, Xinchao Wang
The Kolmogorov-Arnold Transformer replaces MLP layers with Kolmogorov-Arnold Network layers to enhance transformers, overcoming challenges related to inference speed, computation efficiency, and weight initialization through rational basis, group learning, and variance-preserving techniques.
Gilles Baechler, Srinivas Sunkara, Maria Wang +7 authors
ScreenAI, a vision-language model optimized with a flexible patching strategy, achieves state-of-the-art performance on UI and infographic tasks with 5B parameters.
Xiaochuang Han, Marjan Ghazvininejad, Pang Wei Koh +1 authors
Using canonical codec representations (JPEG, AVC) as a direct modeling approach for image and video generation outperforms pixel-based and vector quantization methods, facilitating multi-modal LLMs.
Viraat Aryabumi, Yixuan Su, Raymond Ma +6 authors
The inclusion of code data during pre-training significantly improves LLMs' performance across various natural language and world knowledge tasks, beyond code generation.
Sainbayar Sukhbaatar, Olga Golovneva, Vasu Sharma +8 authors
Branch-Train-MiX (BTX) method enhances Large Language Models by asynchronously training experts in parallel and integrating them using Mixture-of-Expert layers with token-level routing for improved accuracy and efficiency.
Jun Zhan, Junqi Dai, Jiasheng Ye +13 authors
AnyGPT is a multimodal language model using discrete representations to process and generate content across various modalities with performance on par with specialized models.
Yew Ken Chia, Liying Cheng, Hou Pong Chan +5 authors
A benchmark and framework named M-LongDoc evaluates and improves large multimodal models for document reading, focusing on open-ended solutions and retrieval tasks over lengthy documents.
Yi Lu, Jing Nathan Yan, Songlin Yang +6 authors
A controlled evaluation protocol for long-context models reveals that exact fine-tuning methods are effective, while approximate attention methods underperform, and extrapolation remains challenging.
Abhimanyu Hans, Avi Schwarzschild, Valeriia Cherepanova +5 authors
A novel LLM detector using contrasting scores from two pre-trained models achieves high accuracy in identifying machine-generated text without any training data.
Liqun Yang, Jian Yang, Chaoren Wei +13 authors
A framework uses fine-tuned large language models to guide input mutation in fuzzing, improving the detection of software vulnerabilities.
Boyuan Chen, Diego Marti Monso, Yilun Du +3 authors
Diffusion Forcing trains diffusion models to denoise tokens with independent noise levels, combining next-token prediction and full-sequence diffusion strengths for sequence generation and decision-making tasks.
Xiwei Hu, Rui Wang, Yixiao Fang +3 authors
ELLA, an Efficient Large Language Model Adapter, enhances text-to-image diffusion models by integrating powerful Large Language Models through a Timestep-Aware Semantic Connector, improving dense prompt comprehension and generation quality.
Kaibing Chen, Dong Shen, Hanwen Zhong +14 authors
A multi-modal language model using cross-attention, hierarchical ViT features, and Mixture of Experts mechanism achieves competitive performance in image and video captioning tasks with reduced computational costs.
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.
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.
Yinghao Ma, Anders Øland, Anton Ragni +40 authors
A review of foundation models in music, including large language models and latent diffusion models, highlights their impact, architectural choices, and the need for ethical considerations in music applications.
Xuchen Pan, Dawei Gao, Yuexiang Xie +5 authors
Enhancements to the AgentScope platform improve scalability, efficiency, and ease of use for large-scale multi-agent simulations through distributed mechanisms, flexible environments, and user-friendly tools.
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.
Alon Ziv, Itai Gat, Gael Le Lan +6 authors
MAGNeT is a single-stage, non-autoregressive transformer for audio generation that uses masked spans during training, a rescoring method, and a hybrid approach to achieve fast generation.
Wilfried Bounsi, Borja Ibarz, Andrew Dudzik +5 authors
A novel TransNAR model combines Transformer-based language understanding with graph neural network solvers to enhance algorithmic reasoning.
Mo Li, Songyang Zhang, Yunxin Liu +1 authors
NeedleBench and the Ancestral Trace Challenge evaluate the long-context retrieval and reasoning capabilities of large language models, highlighting their limitations in practical applications.
Fengji Zhang, Linquan Wu, Huiyu Bai +6 authors
HumanEval-V provides a comprehensive benchmark for evaluating LMMs' diagram interpretation and reasoning abilities, particularly in coding contexts, revealing significant areas for improvement.
Ting Jiang, Minghui Song, Zihan Zhang +6 authors
E5-V, a novel framework, utilizes multimodal large language models (MLLMs) to achieve universal multimodal embeddings with strong performance and reduced costs through single modality training.
Runpeng Yu, Weihao Yu, Xinchao Wang
A comprehensive comparison of KAN and MLP models across diverse tasks reveals that MLP generally outperforms KAN except in symbolic formula representation where KAN's B-spline activation function provides advantage, and KAN exhibits more severe forgetting issues in class-incremental continual learning.
Zehui Chen, Kuikun Liu, Qiuchen Wang +4 authors
MindSearch, an LLM-based multi-agent framework, improves web information seeking and integration through parallel processing and hierarchical retrieval, achieving better performance than existing solutions.
Aleksei Bochkovskii, Amaël Delaunoy, Hugo Germain +4 authors
Depth Pro is a fast and accurate zero-shot monocular depth estimation model that generates high-resolution depth maps using a multi-scale transformer and a combined real-synthetic training protocol.
Di Liu, Meng Chen, Baotong Lu +11 authors
RetrievalAttention is a training-free method that accelerates attention computation in Transformer-based LLMs by using attention-aware vector search to reduce retrieval costs and GPU memory usage.
Shihan Dou, Yan Liu, Haoxiang Jia +13 authors
StepCoder, a reinforcement learning framework, enhances code generation by breaking tasks into subtasks and optimizing only executed code segments, using the APPS+ dataset for training.
Noveen Sachdeva, Benjamin Coleman, Wang-Cheng Kang +6 authors
Data-efficient methods like Ask-LLM and Density sampling improve model quality and training efficiency in large language models by optimizing data selection and coverage.
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