Hybrid convolutional neural network yields energy-efficient approach with analog first layer

Oscillatory retinal neuron networks don’t require external voltage sources and show comparable performance to cutting-edge GPU-based convolutional neural networks, for energy costs thousands of times less than current approaches.