In a paper published in the journal Nature Optics on the 12th, scientists at the Massachusetts Institute of Technology (MIT) said they have developed a completely new optical neural network system that can perform highly complex computations and greatly enhance the "deep learning "System operation speed and efficiency.
a, general artificial neural network architecture by the input layer, a plurality of hidden layer and the output layer. b, decompose the general neural network into a single layer. c, constitute each layer of artificial neural network optical interference and nonlinear units. d, proposed an all-optical, fully integrated neural network.
"Deep learning" system simulates the learning ability of the human brain through artificial neural network and has become a hot research field in computer field. However, due to the highly complex computations required to perform a large number of repetitive "matrix multiplications" in analogue neural network tasks, such computations can be performed on conventional CPUs or GPU chips running on electricity Too dense, it is very "difficult" to finish.
a, schematic diagram of two ONN experiments. b, experimental feedback and control loop used in the experiment. c. Experiments show that the optical micrographs of the OIU demonstrate that its complete optics enable matrix multiplication (highlighted in red) and attenuation (highlighted in blue). d, schematic diagram of a single phase shifter in MZI and transfer curve for tuning the internal phase shifter.
Environmental Equipment,Environmental Protection Equipment,Pure Water Equipment,Waste Filtration Device
Wuxi Xingyi Intelligent Environment Protection Equipment Co.,Ltd. , https://www.wxxingyi.com