Efficient Decoupled Neural Architecture Search by Structure and Operation Sampling

TL;DR

This paper introduces a decoupled neural architecture search method using reinforcement learning that independently samples structures and operations, significantly improving efficiency while maintaining competitive accuracy.

Contribution

It presents a novel decoupled search algorithm that enhances efficiency and interpretability in neural architecture search compared to traditional RNN controller-based methods.

Findings

Achieves competitive accuracy with reduced search cost

Provides interpretable policy vectors during training

Outperforms state-of-the-art methods in efficiency

Abstract

We propose a novel neural architecture search algorithm via reinforcement learning by decoupling structure and operation search processes. Our approach samples candidate models from the multinomial distribution on the policy vectors defined on the two search spaces independently. The proposed technique improves the efficiency of architecture search process significantly compared to the conventional methods based on reinforcement learning with the RNN controllers while achieving competitive accuracy and model size in target tasks. Our policy vectors are easily interpretable throughout the training procedure, which allows to analyze the search progress and the discovered architectures; the black-box characteristics of the RNN controllers hamper understanding training progress in terms of policy parameter updates. Our experiments demonstrate outstanding performance compared to the state-of-the-art methods with a fraction of search cost.