"AlphaGo" beat Li Shishi 4:1, which made artificial intelligence pay more attention. AlphaGo currently uses about 170 graphics processing units (GPUs) and 1200 central processing units (CPUs), which may require a single computer room, high-power air conditioning, and multiple experts for system maintenance. In this regard, Chinese scientists claim that the number of chips currently used by AlphaGo will be replaced by a chip of the "Cambrian" architecture developed by the Chinese in the future. It is estimated that a small box will be fully loaded. This means that the "Alpha Dog" will be able to run faster.

The reporter learned from the Institute of Computing Technology of the Chinese Academy of Sciences on the 15th that the research team Chen Yunqi and Chen Tianshi first proposed the deep learning processor architecture Cambrian in the world. This month, their deep learning processor instruction set, DianNaoYu, was accepted by ISCA2016, the top international conference in computer architecture, and ranked first in nearly 300 submissions. The first author of the paper is Dr. Liu Shaoli.

AlphaGo is based on deep learning technology to defeat Li Shishi. Deep learning is a kind of intelligent processing technology developed by reference to biological multi-layer neural network processing mode. It has been widely used in computer vision, speech recognition, natural language processing, and audio recognition by Microsoft, Google, Facebook, Ali, Baidu and other companies. And other fields such as bioinformatics.

"But the basic operation of deep learning is the processing of neurons and synapses. The traditional processor instruction set (X86 and ARM, etc.) is developed for general-purpose computing. The basic operations are arithmetic operations and logical operations, which are often required. Hundreds or even thousands of instructions can complete a neuron's processing, and deep learning is not efficient." Chen Tianshi said that Google even needs to use tens of thousands of X86 CPU cores to run for 7 days to train a cat's face. Learn neural networks.

Chen Yunqi and Chen Tianshi's deep learning processor instruction set - DianNaoYu directly faces the processing of large-scale neurons and synapses, and a set of neurons can be processed by one instruction. Chen Tianshi explained that, unlike the general-purpose processor that Google uses, the processor chip we designed is specifically for deep learning technology. “Universal processors do deep learning like cutting meat with a Swiss army knife, and we design a special meat cleaver to cut.”

Concerned about surprises

Christmas Lights

Christmas Lights,Christmas Led Pixel Light,Rgb Led String Lights,Pixel Decoration Light

Shen zhen SH LED Technology Co.,Ltd , https://www.pixellightsolutions.com