RESUMEN
Deep-learning models have become pervasive tools in science and engineering. However, their energy requirements now increasingly limit their scalability1. Deep-learning accelerators2-9 aim to perform deep learning energy-efficiently, usually targeting the inference phase and often by exploiting physical substrates beyond conventional electronics. Approaches so far10-22 have been unable to apply the backpropagation algorithm to train unconventional novel hardware in situ. The advantages of backpropagation have made it the de facto training method for large-scale neural networks, so this deficiency constitutes a major impediment. Here we introduce a hybrid in situ-in silico algorithm, called physics-aware training, that applies backpropagation to train controllable physical systems. Just as deep learning realizes computations with deep neural networks made from layers of mathematical functions, our approach allows us to train deep physical neural networks made from layers of controllable physical systems, even when the physical layers lack any mathematical isomorphism to conventional artificial neural network layers. To demonstrate the universality of our approach, we train diverse physical neural networks based on optics, mechanics and electronics to experimentally perform audio and image classification tasks. Physics-aware training combines the scalability of backpropagation with the automatic mitigation of imperfections and noise achievable with in situ algorithms. Physical neural networks have the potential to perform machine learning faster and more energy-efficiently than conventional electronic processors and, more broadly, can endow physical systems with automatically designed physical functionalities, for example, for robotics23-26, materials27-29 and smart sensors30-32.
RESUMEN
The production effect-that reading aloud leads to better memory than does reading silently-has been defined narrowly with reference to memory; it has been explored largely using word lists as the material to be read and remembered. But might the benefit of production extend beyond memory and beyond individual words? In a series of four experiments, passages from reading comprehension tests served as the study material. Participants read some passages aloud and others silently. After each passage, they completed multiple-choice questions about that passage. Separating the multiple-choice questions into memory-focused versus comprehension-focused questions, we observed a consistent production benefit only for the memory-focused questions. Production clearly improves memory for text, not just for individual words, and also extends to multiple-choice testing. The overall pattern of findings fits with the distinctiveness account of production-that information read aloud stands out at study and at test from information read silently. Only when the tested information is a very close match to the studied information, as is the case for memory questions but not for comprehension questions, does production improve accuracy.