NAS-BNN: Neural Architecture Search for Binary Neural Networks

Zhihao Lin, Yongtao Wang, Jinhe Zhang, Xiaojie Chu, Haibin Ling

Abstract: Highlights•A novel search space for binary neural networks is proposed.•Non-Decreasing (ND) constraint is proposed to remove the ineffective subnets, building a more powerful search space.•Three training techniques are proposed to enhance the training process of binary supernet, so as to boost the performance of each subnet.•Extensive experiments demonstrate that our method achieves better trade-offs than prior binary neural networks.