Neural Architecture Search by Learning a Hierarchical Search Space

Mehraveh Javan Roshtkhari, Matthew Toews, Marco Pedersoli

Keywords: Neural Architecture Search, Monte-Carlo Tree Search, Hierarchical Search Space, Hierarchical Clustering

TL;DR: We propose a new method to learn hierarchical structure of NAS space for more efficient Monte-Carlo Tree Search method.

Abstract:

Monte-Carlo Tree Search (MCTS) is a powerful tool for many non-differentiable search related problems such as adversarial games. However, the performance of such approach highly depends on the order of the nodes that are considered at each branching of the tree. If the first branches are not discriminative enough, i.e. they cannot distinguish between promising and deceiving configurations for the final task, the efficiency of the search is exponentially reduced. While in some cases the order of the branching is given as part of the problem (e.g. in chess the sequential order of the moves is defined by the game), in others, such as Neural Architecture Search (NAS), the visiting order of the tree is not important, and only the final architecture matters. In this paper, we study the application of MCTS to NAS for the task of image classification. We analyze several sampling methods and branching alternatives for MCTS and propose to learn the branching by hierarchical clustering of architectures based on their similarity. The similarity is measured by the pairwise distance of output vectors of architectures. Extensive experiments on two challenging benchmarks on CIFAR10 and ImageNet show that MCTS, if provided with a good branching hierarchy, can yield promising solutions more efficiently than other approaches for NAS problems.

Primary Area: probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 11616