Hybrid Composition with IdleBlock: More Efficient Networks for Image Recognition

TL;DR

This paper introduces IdleBlock, a new network building block that prunes connections, and demonstrates that hybrid composition with IdleBlock significantly improves the efficiency and accuracy of MobileNet v3 and EfficientNet-B0 for image recognition.

Contribution

The paper presents IdleBlock, a novel pruning-based building block, and shows that hybrid composition with IdleBlock enhances existing efficient networks without neural architecture search.

Findings

Deeper MobileNet v3 with IdleBlock surpasses state-of-the-art networks.

Hybrid EfficientNet-B0 networks are more efficient at similar computation budgets.

Results suggest a new direction for simpler, more efficient network design.

Abstract

We propose a new building block, IdleBlock, which naturally prunes connections within the block. To fully utilize the IdleBlock we break the tradition of monotonic design in state-of-the-art networks, and introducing hybrid composition with IdleBlock. We study hybrid composition on MobileNet v3 and EfficientNet-B0, two of the most efficient networks. Without any neural architecture search, the deeper "MobileNet v3" with hybrid composition design surpasses possibly all state-of-the-art image recognition network designed by human experts or neural architecture search algorithms. Similarly, the hybridized EfficientNet-B0 networks are more efficient than previous state-of-the-art networks with similar computation budgets. These results suggest a new simpler and more efficient direction for network design and neural architecture search.