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。
Hongyu Wang, Shuming Ma, Li Dong +7 authors
BitNet, a 1-bit Transformer architecture, reduces memory and energy consumption while achieving competitive performance in language modeling.
42 篇论文 · 按点赞排序
Akari Asai, Zeqiu Wu, Yizhong Wang +2 authors
Self-Reflective Retrieval-Augmented Generation (Self-RAG) enhances large language models by adaptively retrieving and reflecting on passages, improving factuality and performance across various tasks.
Chunyi Sun, Junlin Han, Weijian Deng +3 authors
3D-GPT leverages large language models to drive 3D modeling tasks, enhancing scene descriptions and interfacing with 3D software through integrated agents.
Zhangir Azerbayev, Hailey Schoelkopf, Keiran Paster +6 authors
Llemma, a large language model pretrained on mathematical data, outperforms existing models and demonstrates tool use and formal theorem proving capabilities.
Peng Li, Yeye He, Dror Yashar +6 authors
A new table-tuning paradigm improves language models' understanding and generalization in table-related tasks by fine-tuning them with synthesized table data, resulting in enhanced performance.
Zhen Xu, Sida Peng, Haotong Lin +5 authors
4K4D, a 4D point cloud representation with a hybrid appearance model and differentiable depth peeling algorithm, achieves high-speed and high-quality dynamic view synthesis at 4K resolution.
Aohan Zeng, Mingdao Liu, Rui Lu +4 authors
AgentTuning enhances the agent capabilities of open LLMs through instruction-tuning without compromising their general abilities, demonstrating performance comparable to commercial models on unseen tasks.
Dawid Jan Kopiczko, Tijmen Blankevoort, Yuki Markus Asano
Vector-based Random Matrix Adaptation (VeRA) reduces the number of trainable parameters by 10x compared to LoRA while maintaining performance, and is demonstrated on benchmarks like GLUE and E2E, showing its utility in instruction-following.
Yixiao Li, Yifan Yu, Chen Liang +4 authors
LoftQ, a novel quantization framework, improves performance of LLMs by simultaneously quantizing and initializing LoRA fine-tuning, reducing the gap in downstream task performance.
Jianwei Yang, Hao Zhang, Feng Li +3 authors
Set-of-Mark (SoM) enhances visual grounding in large multimodal models using segmented and marked images, demonstrating superior performance on fine-grained vision tasks without full fine-tuning.
Weijia Shi, Sewon Min, Maria Lomeli +7 authors
In-Context Pretraining improves large language models by ordering documents to maximize contextual similarity, leading to better performance in complex reasoning tasks.
Xi Chen, Xiao Wang, Lucas Beyer +16 authors
PaLI-3, a smaller vision language model, outperforms larger models on multimodal tasks, particularly localization and text understanding, using SigLIP pretraining and achieves state-of-the-art results in multilingual cross-modal retrieval.
Josef Dai, Xuehai Pan, Ruiyang Sun +5 authors
Safe RLHF, a novel algorithm for human value alignment in LLMs, improves performance and safety by decoupling human preferences and using a Lagrangian method to balance reward and cost constraints.
Yecheng Jason Ma, William Liang, Guanzhi Wang +6 authors
Eureka, an LLM-powered reward design algorithm, generates human-level reward functions enhancing reinforcement learning for complex manipulation tasks and RLHF, including pen spinning with a simulated Shadow Hand.
Dingyao Yu, Kaitao Song, Peiling Lu +5 authors
MusicAgent is a system that integrates various music-related tools and uses large language models to automate music processing tasks based on user requests.
Jun Chen, Deyao Zhu, Xiaoqian Shen +7 authors
MiniGPT-v2 is a unified vision-language model that uses task-specific identifiers to perform diverse tasks with high performance.
Juan Rocamonde, Victoriano Montesinos, Elvis Nava +2 authors
Pretrained vision-language models can serve as zero-shot reward models in reinforcement learning, enabling complex task learning from minimal natural language prompts.
Yaofang Liu, Xiaodong Cun, Xuebo Liu +7 authors
A new framework evaluates vision-language generative models for video creation by using a nuanced benchmark across visual, content, motion, and textual quality metrics, aligned with user opinions.
Christopher Fifty, Dennis Duan, Ronald G. Junkins +4 authors
A meta-learning algorithm learns new visual concepts during inference without fine-tuning, leveraging a pre-trained feature extractor and sequence modeling.
Tao Li, Gang Li, Zhiwei Deng +2 authors
A zero-shot agent learns to perform tasks in a computer environment without expert traces, using self-reflection and structured thought to improve performance, outperforming recent models on easy tasks and matching them on more complex ones.
Aman Madaan, Pranjal Aggarwal, Ankit Anand +10 authors
AutoMix strategically routes queries to larger LLMs based on a noise-refined self-verification mechanism, improving the incremental benefit per cost.
Yixiao Zhang, Akira Maezawa, Gus Xia +2 authors
Loop Copilot, a multi-round dialogue system using a large language model, facilitates music creation by iteratively refining and selecting specialized AI models for task execution while maintaining musical coherence.
Athul Paul Jacob, Yikang Shen, Gabriele Farina +1 authors
A new game-theoretic decoding method, EQUILIBRIUM-RANKING, improves performance across various text generation tasks by reconciling incompatible scoring procedures.
Chen Jin, Ryutaro Tanno, Amrutha Saseendran +2 authors
A novel Multi-Concept Prompt Learning framework (MCPL) learns multiple embeddings from a single sentence-image pair to represent object-level concepts in images, using regularisation techniques to enhance accuracy.
Eric Mitchell, Rafael Rafailov, Archit Sharma +2 authors
Emulated fine-tuning (EFT) enables decoupling and adjusting knowledge and skills from pre-training and fine-tuning stages, demonstrating improvements in helpfulness and factuality for large language models.
Ethan Callanan, Amarachi Mbakwe, Antony Papadimitriou +6 authors
LLMs, evaluated using CFA exam questions, show mixed performance in financial reasoning across different scenarios, highlighting areas for improvement and future research.
Boyi Li, Philipp Wu, Pieter Abbeel +1 authors
A framework uses language models for interactive task planning, allowing generalization to new goals with simple task guidelines and precise replanning.
Hung Le, Hailin Chen, Amrita Saha +3 authors
CodeChain improves the modularity and correctness of solutions generated by Large Language Models (LLMs) through a chain of self-revisions and sub-module reuse, achieving significant performance gains on competitive programming tasks.
Xinhua Cheng, Tianyu Yang, Jianan Wang +4 authors
A Progressive3D framework uses locally progressive editing and semantic component suppression to generate precise 3D content from complex text prompts, improving text-to-3D generation accuracy.
Róbert Csordás, Kazuki Irie, Jürgen Schmidhuber
Novel perspectives and methods for MoEs unify and enhance sparse approximations of NNs, improving efficiency without performance loss.
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