BriefGPT.xyz
Feb, 2024
通过凸优化对基于神经网络的生成扩散模型进行分析
Analyzing Neural Network-Based Generative Diffusion Models through Convex Optimization
HTML
PDF
Fangzhao Zhang, Mert Pilanci
TL;DR
本研究提出了一个理论框架,通过将评分匹配和去噪评分匹配视为凸优化问题,对基于两层神经网络的扩散模型进行了分析。尽管现有的扩散理论主要是渐近的,但我们对有限数据的神经网络扩散模型进行了确切的预测评分函数表征,并建立了收敛结果,从而有助于理解非渐近设置中神经网络扩散模型的学习过程。
Abstract
diffusion models
are becoming widely used in state-of-the-art image, video and audio generation. Score-based
diffusion models
stand out among these methods, necessitating the estimation of score function of the i
→