Scholarlink

In this chapter, we describe the flow-based generative models. The core idea is to start from the existing basic probability distribution, transform the probability distribution through a series of deep network layers, and finally define a large class of very broad probability distributions. In this way, we can compute the data distribution with close-form efficient solutions. Based on this idea, we will describe several instantiations, including normalizing flow, RealNVP, and Glow. We will also introduce flow-based model for text representation learning.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Flow-Based Generative Model

  • 工具书词条

  • 发表日期：

    2026年

查看全文
  • Lei Li

摘要

In this chapter, we describe the flow-based generative models. The core idea is to start from the existing basic probability distribution, transform the probability distribution through a series of deep network layers, and finally define a large class of very broad probability distributions. In this way, we can compute the data distribution with close-form efficient solutions. Based on this idea, we will describe several instantiations, including normalizing flow, RealNVP, and Glow. We will also introduce flow-based model for text representation learning.