Towards Accurate and Compact Architectures via Neural Architecture Transformer

TL;DR

This paper introduces NAT and NAT++, methods that optimize neural network architectures by replacing redundant operations with more efficient ones, leading to more accurate and compact models without extra computational cost.

Contribution

The paper proposes NAT++ which enlarges the transition space for architecture optimization using a two-level rule and BMSoftmax, improving performance over previous methods.

Findings

Transformed architectures outperform original and existing optimized models.

NAT++ achieves higher accuracy with fewer parameters.

The approach effectively reduces redundancy in neural architectures.

Abstract

Designing effective architectures is one of the key factors behind the success of deep neural networks. Existing deep architectures are either manually designed or automatically searched by some Neural Architecture Search (NAS) methods. However, even a well-designed/searched architecture may still contain many nonsignificant or redundant modules/operations. Thus, it is necessary to optimize the operations inside an architecture to improve the performance without introducing extra computational cost. To this end, we have proposed a Neural Architecture Transformer (NAT) method which casts the optimization problem into a Markov Decision Process (MDP) and seeks to replace the redundant operations with more efficient operations, such as skip or null connection. Note that NAT only considers a small number of possible transitions and thus comes with a limited search/transition space. As a result, such a small search space may hamper the performance of architecture optimization. To address this issue, we propose a Neural Architecture Transformer++ (NAT++) method which further enlarges the set of candidate transitions to improve the performance of architecture optimization. Specifically, we present a two-level transition rule to obtain valid transitions, i.e., allowing operations to have more efficient types (e.g., convolution->separable convolution) or smaller kernel sizes (e.g., 5x5->3x3). Note that different operations may have different valid transitions. We further propose a Binary-Masked Softmax (BMSoftmax) layer to omit the possible invalid transitions. Extensive experiments on several benchmark datasets show that the transformed architecture significantly outperforms both its original counterpart and the architectures optimized by existing methods.