Large Language Models as Optimizers
Chengrun Yang, Xuezhi Wang, Yifeng Lu +4 authors
OPRO, a method using large language models to optimize tasks described in natural language, outperforms human-designed prompts on various benchmark datasets.
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。
本周最热87 赞
YaRN: Efficient Context Window Extension of Large Language Models
Bowen Peng, Jeffrey Quesnelle, Honglu Fan +1 authors
YaRN extends the context window of transformer-based language models like LLaMA with improved efficiency and performance.
42 篇论文 · 按点赞排序
RLAIF: Scaling Reinforcement Learning from Human Feedback with AI Feedback
Harrison Lee, Samrat Phatale, Hassan Mansoor +5 authors
Reinforcement learning from AI feedback (RLAIF) achieves similar performance to reinforcement learning from human feedback (RLHF) in aligning large language models with human preferences, offering a scalable alternative.
FLM-101B: An Open LLM and How to Train It with $100K Budget
Xiang Li, Yiqun Yao, Xin Jiang +10 authors
A growth strategy reduces the cost of training large language models while achieving comparable performance, with a focus on evaluating IQ and minimizing memorization.
DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models
Yung-Sung Chuang, Yujia Xie, Hongyin Luo +3 authors
A decoding strategy that contrasts logits from different transformer layers reduces hallucinations in large language models without external knowledge or fine-tuning.
One Wide Feedforward is All You Need
Telmo Pessoa Pires, António V. Lopes, Yannick Assogba +1 authors
The study shows that the Feed Forward Network in Transformers is highly redundant and its reduction or sharing within the model can lead to improved accuracy and latency without significant loss of performance.
SLiMe: Segment Like Me
Aliasghar Khani, Saeid Asgari Taghanaki, Aditya Sanghi +2 authors
SLiMe segments images at desired granularity using Stable Diffusion with minimal annotations and outperforms existing one-shot and few-shot segmentation methods.
Tracking Anything with Decoupled Video Segmentation
Ho Kei Cheng, Seoung Wug Oh, Brian Price +2 authors
DEVA, a decoupled approach using image-level segmentation and bi-directional temporal propagation, achieves favorable results in various data-scarce video segmentation tasks with reduced annotation and training costs.
ProPainter: Improving Propagation and Transformer for Video Inpainting
Shangchen Zhou, Chongyi Li, Kelvin C. K. Chan +1 authors
ProPainter is a framework that improves video inpainting by combining dual-domain propagation and a mask-guided sparse video Transformer, outperforming existing methods in PSNR while maintaining efficiency.
AniPortraitGAN: Animatable 3D Portrait Generation from 2D Image Collections
Yue Wu, Sicheng Xu, Jianfeng Xiang +4 authors
The proposed 3D-aware GAN generates high-quality, controllable portrait images from unstructured 2D images, addressing challenges in facial and head pose representation.
VideoGen: A Reference-Guided Latent Diffusion Approach for High Definition Text-to-Video Generation
Xin Li, Wenqing Chu, Ye Wu +7 authors
VideoGen uses text-to-image models and latent diffusion to generate high-definition, temporally consistent videos from text prompts, achieving state-of-the-art performance.
Large Content And Behavior Models To Understand, Simulate, And Optimize Content And Behavior
Ashmit Khandelwal, Aditya Agrawal, Aanisha Bhattacharyya +8 authors
Large Content and Behavior Models (LCBMs) are introduced to address the third level of communication by incorporating behavior tokens into LLM training, demonstrating improved generalization on behavior simulation and domain adaptation.
ModelScope-Agent: Building Your Customizable Agent System with Open-source Large Language Models
Chenliang Li, Hehong Chen, Ming Yan +11 authors
A customizable agent framework, ModelScope-Agent, equips large language models with tool-use capabilities for real-world applications by integrating them with extensive external APIs and supporting model training on multiple open-source models.
CityDreamer: Compositional Generative Model of Unbounded 3D Cities
Haozhe Xie, Zhaoxi Chen, Fangzhou Hong +1 authors
CityDreamer is a compositional generative model that excels in generating realistic 3D cities by separating building generation from background objects using distinct modules and leveraging real-world datasets.
FACET: Fairness in Computer Vision Evaluation Benchmark
Laura Gustafson, Chloe Rolland, Nikhila Ravi +5 authors
A benchmark named FACET is introduced to evaluate performance disparities of computer vision models across demographic attributes through extensive annotations and intersectional analysis.
GPT Can Solve Mathematical Problems Without a Calculator
Zhen Yang, Ming Ding, Qingsong Lv +5 authors
A large language model with sufficient training data can accurately perform complex arithmetic operations without calculator tools, outperforming GPT-4.
ImageBind-LLM: Multi-modality Instruction Tuning
Jiaming Han, Renrui Zhang, Wenqi Shao +14 authors
ImageBind-LLM uses a learnable bind network to enable large language models to follow multi-modal instructions through image-text alignment and a visual cache model.
ControlMat: A Controlled Generative Approach to Material Capture
Giuseppe Vecchio, Rosalie Martin, Arthur Roullier +4 authors
A method called ControlMat uses a diffusion model to generate high-resolution, tileable digital materials from a single photograph, addressing challenges of uncontrolled lighting.
Scaling Autoregressive Multi-Modal Models: Pretraining and Instruction Tuning
Lili Yu, Bowen Shi, Ramakanth Pasunuru +24 authors
CM3Leon, a multi-modal language model, achieves state-of-the-art text-to-image generation with efficient training and advanced control over image generation tasks.
Efficient RLHF: Reducing the Memory Usage of PPO
Michael Santacroce, Yadong Lu, Han Yu +2 authors
Hydra-RLHF optimizes Reinforcement Learning with Human Feedback by integrating SFT and Reward models and dynamically disabling LoRA, reducing memory usage and latency while maintaining performance.
Matcha-TTS: A fast TTS architecture with conditional flow matching
Shivam Mehta, Ruibo Tu, Jonas Beskow +2 authors
Matcha-TTS is a new encoder-decoder architecture for text-to-speech that uses optimal-transport conditional flow matching to produce high-quality outputs quickly and efficiently, outperforming existing models in speed, memory usage, and audio quality.
PromptTTS 2: Describing and Generating Voices with Text Prompt
Yichong Leng, Zhifang Guo, Kai Shen +12 authors
PromptTTS 2 addresses the challenges in text-to-speech systems by using a variation network to capture voice variability not expressed in text prompts and a prompt generation pipeline utilizing large language models to create high-quality text prompts.
InstructDiffusion: A Generalist Modeling Interface for Vision Tasks
Zigang Geng, Binxin Yang, Tiankai Hang +8 authors
InstructDiffusion is a diffusion-based framework that interprets human instructions to manipulate images, achieving versatility across vision tasks and novel datasets.
SyncDreamer: Generating Multiview-consistent Images from a Single-view Image
Yuan Liu, Cheng Lin, Zijiao Zeng +4 authors
SyncDreamer generates multiview-consistent images using a synchronized diffusion model and 3D-aware feature attention mechanism, improving the quality of 3D generation tasks.
Hierarchical Masked 3D Diffusion Model for Video Outpainting
Fanda Fan, Chaoxu Guo, Litong Gong +5 authors
A masked 3D diffusion model with cross-attention and a hybrid coarse-to-fine pipeline achieves state-of-the-art results in video outpainting by maintaining temporal consistency and reducing artifacts.
Point-Bind & Point-LLM: Aligning Point Cloud with Multi-modality for 3D Understanding, Generation, and Instruction Following
Ziyu Guo, Renrui Zhang, Xiangyang Zhu +8 authors
Point-Bind aligns 3D point clouds with various modalities using ImageBind, enabling applications like 3D generation and understanding, while Point-LLM enhances pre-trained LLMs for 3D instructions via parameter-efficient techniques.
Text2Control3D: Controllable 3D Avatar Generation in Neural Radiance Fields using Geometry-Guided Text-to-Image Diffusion Model
Sungwon Hwang, Junha Hyung, Jaegul Choo
Text2Control3D generates controllable 3D avatars from monocular videos using diffusion models, Neural Radiance Fields, and cross-reference attention.
StyleAdapter: A Single-Pass LoRA-Free Model for Stylized Image Generation
Zhouxia Wang, Xintao Wang, Liangbin Xie +4 authors
StyleAdapter introduces a method for generating high-quality, stylized images in a single pass by processing prompt and style reference features separately within a unified model.
Doppelgangers: Learning to Disambiguate Images of Similar Structures
Ruojin Cai, Joseph Tung, Qianqian Wang +3 authors
A learning-based approach using a custom dataset and network architecture distinguishes illusory image matches, improving 3D reconstruction accuracy.
Gated recurrent neural networks discover attention
Nicolas Zucchet, Seijin Kobayashi, Yassir Akram +4 authors
Recent RNNs with linear recurrent layers and multiplicative gating can implement linear self-attention, similar to Transformers, as discovered through reverse-engineering of trained RNNs on in-context learning tasks.