Neural Architecture Search: A Survey
Users: 1 - Average Rating: 4.00
The article is about automated neural architecture search (NAS) methods. Automated search methods have arisen attention because manually adjusting of parameters by human experts, which is the current practice, is a time-consuming and prone to errors process. This article reviews the existing work in this field of research, dividing the material according to three dimensions, which are the three key steps of NAS methods: search space (defining which architectures can be represented), search strategy (detailing how to explore the search space, often exponentially large or even unbounded), and performance estimation strategy (the process of estimating the performance of each tested architecture).
Type:
Scientific Paper
Area:
Data Analytics, Machine Learning
Target Group:
Advanced
DOI:
Not available
Cite as:
Elsken, T. and Metzen, J. H. and Hutter F., Neural architecture search: A survey, The Journal of Machine Learning Research 20.1 (2019): 1997-2017.
Author of the review:
Giulia Cademartori
University of Genoa
You have to login to leave a comment. If you are not registered click here
Stefano Zampini
Interesting reading if you want to explore how to automatize the design of Neural Network architecture. This is in fact a review of existing works in this field and can be a good starting point to approach this subject!