Practical Block-Wise Neural Network Architecture Generation

TLDR

Convolutional neural networks have achieved remarkable success in computer vision, yet most practical architectures are hand‑crafted and demand expert design. This work introduces BlockQNN, a block‑wise neural network generation pipeline that automatically constructs high‑performance models using Q‑learning with epsilon‑greedy exploration. The method trains a learning agent to sequentially select component layers for an optimal block, stacks these blocks into a full network, and employs a distributed asynchronous framework with early stopping to accelerate search. BlockQNN produces networks that outperform hand‑crafted state‑of‑the‑art models, achieving a 3.54 % top‑1 error on CIFAR‑10, requires only three days on 32 GPUs, and generalizes well to ImageNet.

Abstract

Convolutional neural networks have gained a remarkable success in computer vision. However, most usable network architectures are hand-crafted and usually require expertise and elaborate design. In this paper, we provide a block-wise network generation pipeline called BlockQNN which automatically builds high-performance networks using the Q-Learning paradigm with epsilon-greedy exploration strategy. The optimal network block is constructed by the learning agent which is trained sequentially to choose component layers. We stack the block to construct the whole auto-generated network. To accelerate the generation process, we also propose a distributed asynchronous framework and an early stop strategy. The block-wise generation brings unique advantages: (1) it performs competitive results in comparison to the hand-crafted state-of-the-art networks on image classification, additionally, the best network generated by BlockQNN achieves 3.54% top-1 error rate on CIFAR-10 which beats all existing auto-generate networks. (2) in the meanwhile, it offers tremendous reduction of the search space in designing networks which only spends 3 days with 32 GPUs, and (3) moreover, it has strong generalizability that the network built on CIFAR also performs well on a larger-scale ImageNet dataset.