«Samooborony neuromorphic processor» Intel Loihi is designed to accelerate applications AI

Intel has described the development, which can accelerate applications of artificial intelligence and to expand the scope of their application. This processor under the name of Loihi, which the manufacturer describes as «samooborony» and «neuromorphic».

As stated, the principles of Loihi similar to those which a human brain works. The processor is trained using different types of feedback. As a result, he becomes «smarter» in the process, excludes the usual current AI systems is the training phase. Loihi architecture resembles a neural network, in which the pulses in the synapses serve to create and strengthen the connections between neurons. Intelligent behavior is the result of cooperative and competitive interactions between network sections and the outside world.

Loihi test chip includes digital circuitry that reproduces the basic elements of the brain, allowing you to speed up machine learning while reducing energy consumption. On the one hand, the approach allows the processor to learn autonomously and in real time, without waiting for updates coming from the cloud, and on the other to radically increase the speed. Evaluation of the Intel, training Loihi happens 1 million times faster in comparison with other neural networks, if we compare the total number of operations required to achieve a given accuracy when solving problems of recognition MNIST. Moreover, in comparison with such approaches like the convolutional neural network and neural network deep learning, Loihi spends significantly less resources for the same task. With regard to energy efficiency, Loihi 1000 times greater than the General-purpose CPU used in conventional learning systems.

Processor, made on 14-nanometer technology, includes 130,000 130 million neurons and synapses.

Intel promises in the first half of 2018 to provide Loihi leading universities and research institutions, specializing in the development of AI.

Tags:
Intel

Comment

(Visited 19 times, 1 visits today)
No tags for this post.