Rate-energy-accuracy optimization of convolutional architectures for face recognition.

•We optimize the energy-rate-accuracy characteristics of a convolutional architecture for face recognition.•We profile a convolutional network implementation on low-power hardware, i.e., a Raspberry Pi.•We obtain a speedup of 17 without significantly affecting task accuracy.•We propose a coding architecture tailored to features extracted by such model, obtaining very compact yet effective features.