Editorial: Deep Learning in Biological, Computer, and Neuromorphic Systems

Deep learning is a branch of machine learning in which statistical representations of the input data, as opposed to task-specific algorithms, are learned. Deep learningmay be supervised, unsupervised, or semi-supervised (Lecun et al., 2015; Schmidhuber, 2015; Goodfellow et al., 2016). Deep learning techniques are behind many impressive recent successes of machine learning; for example, a deep learning machine recently beat the world champion at the game of Go (Silver et al., 2016), a highly significant achievement that had remained out of reach for the past 50 years. The current Research Topic provides a useful overview of the capabilities and the applications of deep learning and of attempts to elucidate some of its general principles. Many of the applications in the current Research Topic focus on various aspects of neurology and neuroscience. For example, Liu et al. use deep learning to distinguish patients with Alzheimer’s disease from healthy controls using PET images; Wang and Ke use an artificial neural network (ANN) to identify epileptic seizures in EEG recordings; Hegdé and Bart use deep learning and deep synthesis to generate artificial but naturalistic mammogram images for psychophysics experiments; and Zhang et al. attempt to perform “mind reading” by reconstructing an image being viewed by a subject based on the subject’s MRI recording. Additional papers address the underpinnings of deep learning as an approach. For example, Wan and Song investigate ways to add hints to a network to improve its performance, and Thiele et al. propose a spiking deep network architecture that is suitable for online deep learning. Finally, Bart and Hegdé investigate the explainability of decisions learned in a weakly guided manner, an issue that is relevant to both biological and artificial learning systems.