RSS
Логотип
Баннер в шапке 1
Баннер в шапке 2

Intel Loihi (neyromorfny processor)

Product
Developers: Intel
Date of the premiere of the system: September, 2017
Last Release Date: March, 2020
Technology: Processors

Content

2020: Training in determination of smells

On March 17, 2020 Intel announced that Loihi is capable to distinguish smells now. For this purpose the company  involved a data set about activity of 72 chemical sensors which then the neyromorfny chip could use to classify ten potentially harmful smells in cooperation with scientists of Cornell University. Also the special algorithm providing training of a system was created.

The deposit to a research of scientists consisted of Cornell University that they as biologists who study impact of smells (molecules of substances) on olfactory receptors and transfer of electric impulses to a brain helped to develop algorithms of determination of smells. Intel, in turn, realized an algorithm in the machine code clear to the neyromorfny Loihi processor.

Intel provided the chip defining smells

At the time of the announcement of development the chip is capable to learn 10 smells, including those which proceed from methane, ammonia and acetone. For each of these connections the chip created an individual picture of neural activity.

Intel claims that Loihi from the first attempt recognized smells of a number of substances, harmful to the person. And the processor showed the speed of training in recognition much above, than the competing platforms. So, the structure of Intel needed 3000 times less than the training selections on a class, than to other similar devices.

According to specialists, the mozgopodobny structure of the chip, along with specialized neurobiological algorithms, can become an original electronic nose.[1] 

Intel is not the only company. which tries to train AI to distinguish smells. The Google Brain Team command works with perfumers to connect molecules of smells with the perceived smells, and the Russian researchers use artificial intelligence for determination of deadly gas mixes and to recreate them using machine learning.

2018

Intel earned $1 billion from neyromorfny and other processors for AI

In 2017 the Intel corporation earned $1 billion from neyromorfny and other processors for artificial intelligence. Achievement, important for Intel, was announced by the head of division of Data Center Group Navin Shena (Navin Shenoy) at the Intel Data Centric Innovation Summit conference, taken place on August 8, 2018 in the corporation headquarters in the city of Santa Clara (California, the USA).

Assessment in $1 billion is based on customer information, announced for what tasks processors are bought and also on own calculations of Intel, Naveen Rao, the head of department of Intel on researches in the field of AI explained to Reuters agency. He emphasized that Intel conservatively approached calculations and that real digit, most likely, significantly higher.[2]

In 2017 the Intel corporation earned $1 billion from neyromorfny and other processors for artificial intelligence

In the conditions of stagnation in the PC market Intel places the increasing emphasis on the equipment for data centers which computing powers ensure functioning mobile and online applications, including those that using artificial intelligence technologies allow to recognize images and a voice.

Intel in recent years achieved significant progress in respect of improvement of the AI chips. According to Navin Shenoya, the efficiency of modern Intel CCPs during the work with artificial intelligence is more than 200 times higher, than was a few years ago. Also he noted that Xeon Cascade Lake processors which output is expected later in 2018 cope with such tasks AI as image understanding 11 times better, VentureBeat tells.[3]

Besides, on an action there was a speech about opportunities for growth of DPC-orinetirovannogo of business of Intel. The corporation calculates that the total amount of target market in this sphere by 2022 will reach $200 billion whereas former assessment provided an indicator at the level of $160 billion in 2021.[4]

Open community for development of the neyromorfny Loihi chip

The Intel company announced in February, 2018 that it will organize open community for testing and development of the self-training neyromorfny Loihi chip, and suggested volunteers to provide possible options of development of AI. Order taking for grants will begin[5] in April, 2018[6].

At the professional meeting devoted to neyromorfny calculations in Oregon (Neuro Inspired Computational Elements, NICE) Intel announced achievements in development of neyromorfny data processing and stated that it is going to open for persons interested access for participation in the further work.

The company selected the direction of training at a basis of the principles by which it occurs in a human brain from development of AI: learning process, according to developers, should make information processing, arriving from the environment. The Neyromorfny Loihi chip as reported Intel, during tests showed "absolute functionality, wide working borders and only several bugs". Loihi is capable to distinguish quickly three-dimensional objects from different viewing angles, to master a data set for training for several seconds, at the same time using only a small part of the opportunities.

To continue improvement of neyromorfny data processing, Intel organized community INRC (Intel Neuromorphic Research Community) in which it hopes to combine efforts both scientists, and developers. About several directions are going to conduct work at once. It will be also development of the theory and algorithms, and attempts to adapt the chip to work with data and the systems of "the real world", and search of application for the solution of modern technology tasks.

The company promised participants of community access to test practices and a multi-user system on the basis of Loihi which was called in abbreviated form by NRaaS (Neuromorphic Research as a Service).

Also Intel is going to issue grants to participants of community on the basis of offers of the specific directions of development. Acceptance of preliminary applications will open on April 2, 2018.

2017: Announcement

In September, 2017 Intel provided the neyromorfny Loihi processor using which, according to in the company, artificial intelligence technologies (AI) will become more available.

Intel says that Loihi works by the same principles, as a human brain. The processor containing 130 thousand artificial neurons and 130 million synapses studies, using different types of a feedback. As a result he becomes "cleverer" directly in the course of work, excepting a grade level, usual for the AI present systems.

Intel provided the processor imitating work of a brain

The architecture of Loihi reminds neural network in which impulses in synapses serve for creation and strengthening of ties between neurons. The intellectual behavior becomes result of cooperative and competitive interaction between sections of network and the outside world.

The Loihi processor can become a basis of the trainee and self-trainee of the computer who will consume one thousand times less energy, than the normal PC for which it is necessary for solving of tasks more than performance.

Besides, new Loihi can become something like the coprocessor supplying with functions of artificial intelligence the central processor  of the computing system.

It is noted that  Loihi is developed for such loadings as development and testing of several algorithms with high algorithmic efficiency for the following tasks: for example, planning of a route, compliance restriction, scattered enciphering, studying of dictionaries, dynamic modeling and adaptation.

The new chip will be made using 14-nanometer technology process. The first devices using it will appear in the market in the first half of 2018. At first they will get to  the research institutes and  the universities which are engaged in development of systems of artificial intelligence.[7]

Read Also

Reason instead of programming. There comes the era of the computers capable to thinking

Notes