Thursday , July 29 2021

Intel aims to simplify AI by updating 'Neural Stick'



(image source: Intel)

Intel today unveiled an upgraded Neural Compute Stick that claims to dramatically simplify the development of neural network applications on drones, IoT, and edges.

It may look like a typical USB thumb, but Intel's Intel Neural Computer Stick 2 (NCS2) today at the AI ​​conference in Beijing is more than that. In fact, the small device has a Myriad X (VPU) image processing unit manufactured by Intel's Movidius subsidiary, which controls a neural network used for forecasting applications.

Intel NCS devices are designed to simplify the development of AI technologies on terminal devices. A laptop computer scientist can convert pre-trained deep learning models to the NCS device, which connects to any standard USB 3.0 port.

NCS was designed primarily (but not exclusively) for computer vision and related applications. The neural network operating in NCS allows developers to create applications that can detect and classify objects with a high level of precision.

Intel says that since starting the original NCS in 2017, more than 10,000 developers have used the technology – along with the associated Intel Distribution of OpenVINO toolkit – to create AI applications that address the real challenges.

Intel's NCS2 can bring the AI ​​computer to a drone (image source: Intel)

For example, one of these developers, Peter Ma, has created a program called pure AI water for detecting harmful bacteria in the water, and the BlueScan AI program for detecting skin cancer, according to Intel. The giant chip also reported to chief Christian Haschek, who used NCS to build a Raspberry Pi-based system that scans the internet for illegal child images and reports it to Interpol. Another is Adam Milton-Barker, who used NCS to create a program that translates the American sign language into real-time text.

"The first-generation Intel Neural Compute Stick has triggered a whole community of AI developers in action with a factor and price that did not exist before," said Naveen Rao, Intel vice president and general manager of the AI ​​Products Group. a press release. "We are excited to see what the community is creating with strong improvement to calculate the power that is powered by the new Intel Neural Compute Stick 2."

Intel says NCS2 is eight times faster than the first-generation NCS, allowing customers to create systems that have quicker conclusion procedures or those that incorporate more data points. He also says he has created new optimizations for the OpenVINO toolkit, which will allow developers to write their code once and develop it in many architectures.

Intel made the announcement at the AI ​​DevCon event in Beijing, where more than 1,000 developers are present today and tomorrow. In addition to NCS2, the company is talking about other offers that are part of its CA strategy, such as:

  • The upcoming Cascade Lake Xeon Stage Processor with Optane DC permanent memory and a set of new AI features called Intel DL Boost. Intel expects Lake Cascade to be dispatched later this year and will conquer in 2019.
  • The range of solutions for Vision accelerator design products, including Intel Movidius VPU and Intel Arria 10 FPGA. All accelerator solutions are based on the OpenVINO tool kit.
  • The upcoming Spring Crest (NNP) neural network processor that will be available on the market in 2019 through Intel Nervana. Intel says Nervan's NNP family is using calculation features specifically for deep AI learning, such as dense matrix multiplication and custom interfaces for parallelism.

Related Objects:

Intel's nGraph, a Universal Deep Learning Compiler

New AI chips to give the GPU a path for deep learning money

Intel Details AI Material Strategy for Post-GPU Age


Source link