Slides

This talk includes three parts:
  • First, I introduce the basics of neural architecture search (NAS), including background, motivation, problem definition.
  • Then, I give an overview of NAS researches, organized by research questions, i.e., “what to search”, “how to search”, “what does the search tell us”. I summarize existing solutions to solve these questions.
  • Finally, I introduce my researches that utilize architecture encoding to answer “how to search” along two directions, i.e., accelerate exploration and improve evaluation.
The talk is in Chinese, and the English-version video will soon be available.

Updated: