Training BatchNorm Only in Neural Architecture Search and Beyond

Abstract

This work investigates the usage of batch normalization in neural architecture search (NAS). Specifically, Frankle et al. [22] find that training BatchNorm only can achieve nontrivial performance. Furthermore, Chen et al. [9] claim that training BatchNorm only can speed up the training of the one-shot NAS supernet over ten times. Critically, there is no effort to understand 1) why training BatchNorm only can find the perform-well architectures with the reduced supernet-training time, and 2) what is the difference between the train-BN-only supernet and the standard-train supernet.We begin by showing that the train-BN-only networks converge to the neural tangent kernel regime, obtain the same training dynamics as train all parameters theoretically. Our proof supports the claim to train BatchNorm only on supernet with less training time. Then, we empirically disclose that train-BN-only supernet provides an advantage on convolutions over other operators, cause unfair competition between architectures. This is due to only the convolution operator being attached with BatchNorm. Through experiments, we show that such unfairness makes the search algorithm prone to select models with convolutions. To solve this issue, we introduce fairness in the search space by placing a BatchNorm layer on every operator. However, we observe that the performance predictor in Chen et al. is inapplicable on the new search space. To this end, we propose a novel composite performance indicator to evaluate networks from three perspectives: expressivity, trainability, and uncertainty, derived from the theoretical property of BatchNorm. We demonstrate the effectiveness of our approach on multiple NAS-benchmarks (NAS-Bench-101, NAS-Bench-201) and search spaces (DARTS search space and MobileNet search space).

Make A Long Image Short: Adaptive Token Length for Vision Transformers

“…However, computational efficiency is critical in real-world scenarios, where the executed computation is translated into power consumption or carbon emission. Many works have tried on reducing the computational cost of CNNs via neural architecture search [10,16,25,54,57], knowledge distillation [20,55], dynamic routing [4,13,43,51,56] and pruning [15,18], but how to accelerate the ViT model have been rarely explored.…”