Email Login
Community
News | Research | DeepLearning | Seminar | Q&A

HPE shows a computer intended to emulate the human brain

capture

HPE's Dot Product Engine is a computer system designed to mimic brain activity.

Credit: HPE

Researchers are working on a new type of chip that may act as an intelligent co-processor

 

Intelligent computers that can make decisions like humans may some day be on Hewlett Packard Enterprise's product roadmap.

The company has been showing a prototype computer designed to emulate the way the brain makes calculations. It's based on a new architecture that could define how future computers work.

Top 25 Windows 10 Free Tools

When Windows 10 gets wonky, there's a whole set of activities to try and set things right. If

Read Now

The brain can be seen as an extremely power-efficient biological computer. Brains take in a lot of data related to sights, sounds and smell, which they have to process in parallel without lagging, in terms of computation speed.

HPE's ultimate goal is to create computer chips that can compute quickly and make decisions based on probabilities and associations, much like how the brain operates. The chips will use learning models and algorithms to deliver approximate results that can be used in decision-making.

 

While it will take years for such chips to be commercially available, HPE is testing its brain-like computing model through a prototype system with circuit boards and memory chips. The computer, which was shown for the first time at the Discover conference held recently in Las Vegas, is designed to operate in a way that the brain’s neurons and synapses work.

The brain forms a neural network of trillions of connections, which are called synapses, to communicate. Approximately 100 billion neurons compute in parallel, establishing these connections and transmitting information via electrical and chemical signals. HPE's researchers are keen on bringing the parallelism of brain activity to wired circuitry.

"We're mimicking that architecture of parallel computation using our memristor technology and a specially designed architecture," said Cat Graves, scientific researcher at Hewlett Packard Labs.

Memristors are a new type of storage and memory that could help future A.I. systems understand data and make more use of it. That's different from today's SSDs and DRAM, which just store data. Like with synapses, the learning and retention on memristor circuits are determined by current and data flow characteristics.

 

In brains, data is stored in specific neurons or cells, and calculations for tasks like image or speech recognition take place in those cells. That's the concept behind HPE's architecture for computers intended to mimic brain operations.

"This has the potential to be incredibly more power efficient, save a lot of time, reduce computing complexity and not be clogging up the bandwidth," Graves said.

In traditional computing, data has to leave storage cells for processing via CPU and memory, which can waste valuable computing resources. HPE's architecture is the opposite -- computation takes place in cells where data is stored, like in brains. Then connections are established between cells, much like synapses.

With such a structure, calculations can be highly parallel, Graves said. Such calculations, called "vector matrix multiplications," lie at the heart of computationally intensive algorithms and applications like image filtering, speech recognition and deep-learning systems, Graves said.

HPE researchers took memristors developed in labs and arranged them in a grid-like structure, which was then hooked up to a computer called the Dot Product Engine. The researchers can switch grid setups on the test bed to figure out which configurations work best for different kinds of algorithms.

The researchers managed 8,000 calculations in one clock cycle in one specific memristor setup.

"In a real chip, it'll be faster because all the hardware we have on these boards will be integrated into the chip itself," Graves said.

This new chip won't replace general-purpose processors like GPUs or CPUs. Any kind of computation in a neuromorphic chip is still approximate, based on probabilities, and may not be entirely accurate.

"It's not particularly accurate to the kind of precision level you care about for some kind of applications," Graves said. "For your bank transactions, I wouldn't want my approximate math to be used there."

The chip may act as a co-processor that can bring intelligence to a computer on tasks like image or speech recognition.

HPE's approach is different from that of companies like Qualcomm, which has a software-based approach, and IBM, which relies on a different chip architecture.

It's early days of research for HPE, but Graves is excited for the future of intelligent computers that can process data much like human brains. It'll take a while for HPE to build a chip that can mimic brain activity, however.

"While we don't know all the answers yet, we're starting to glean a few useful insights," Graves said.

 

link: http://www.computerworld.com/article/3084681/computer-hardware/hpe-shows-a-computer-intended-to-emulate-the-human-brain.html

번호 제목 글쓴이 날짜 조회 수
공지 2019 IVPL 송년회 [1] yjchoi 2019.12.20 265
공지 [한국과총]글로벌 현장연수 참가자 모집 GreatMind 2018.04.04 292
공지 Inform Your Students: Global Student Challenge. Over US$2,000 in Prizes! GreatMind 2017.01.12 4343
공지 nVidia Parallel Computing Software Engineer 채용 GreatMind 2016.11.16 717
27 [News] Prof. Lee Kye-Shin has visited to our group.....!!!! file GreatMind 2016.07.05 530
26 [News] Advisory professor Kim has visited to Prof. Kim Ho-seob who is famous and respected in SunMoon university...!!! file GreatMind 2016.07.05 350
25 [Notice] Mr. Park's paper has been accepted in Displays Journal .!!! GreatMind 2016.06.24 177
» HPE shows a computer intended to emulate the human brain.! GreatMind 2016.06.21 169
23 VR survival horror shocks E3 gamers GreatMind 2016.06.21 364
22 Call for Volunteers: IEEE SPS Visibility Initiative...!! [7] GreatMind 2016.06.21 285
21 Mini IoT device records 5 kinds of data GreatMind 2016.06.03 188
20 ACM Elects Vicki Hanson as President GreatMind 2016.05.27 247
19 Facebook plans to boost its translations using neural networks GreatMind 2016.05.27 224
18 Google's 'Magenta' project will see if AIs can truly make art GreatMind 2016.05.27 3311
17 구글 개발자회의 'I/O 2016', 포인트는 'AI' GreatMind 2016.05.21 83
16 [Notice] Mr. Jung H-K and Y_K have visited to our gruop...! file GreatMind 2016.05.19 166
15 [Notice] Prof. Lee Sang-ho visited to our research group (VICL)...!!!! file GreatMind 2016.05.19 109
14 [Notice] Dr. Kalyan and MR. Lee's paper has been accepted in Information Sciences (IF=4.038)...!!! GreatMind 2016.05.15 120
13 Facebook will turn panoramas into “360 Photos” for feed and Gear VR’s 1M users GreatMind 2016.05.13 138
12 [Notice] Mr. Park's paper has been accepted with minor revision in Displays Journal (Elsevier)...!! GreatMind 2016.05.11 838
11 Researchers build 3D smartphone that bends..!!! GreatMind 2016.05.10 126
10 Prof. KIm who is advisory professor of VICL, has visited to research colleagues SunMoon university on 28 April....!!! file GreatMind 2016.05.03 145
9 Drive.ai Brings Deep Learning to Self-Driving Cars..! GreatMind 2016.05.03 175
8 Sookmyung Women's University PRIME 사업 선정..!!! file GreatMind 2016.05.03 101