AI is expected to help the Internet of Things evolve into AIoT

Hou Junyuan, an IEK project manager at the ITRI said that artificial intelligence will bring about changes in all walks of life and thus change the direction of the future. The hardware architecture of traditional artificial intelligence operations mainly includes a central processing unit (CPU), a graphics processor (GPU), an on-site programmable array (FPGA), and the like.

In the past, the cost of artificial intelligence was very expensive, because the algorithm it performed was to simulate human brain computing, which is a massive parallel operation. This is totally different from the existing computer computing model. Currently there are many dedicated artificial intelligence chips on the market, such as Intel's Intel Xeon Phi, NVIDIA's artificial intelligence super chip Tesler P100 GPU, etc., and prices have declined.

Hou Junyuan pointed out that specific areas of dedicated artificial intelligence systems, because of clear application background requirements, profound domain knowledge, and model building calculations are simple and feasible, and at the level of smartness of individual tests, they can now surpass human intelligence and achieve concrete results in many areas. Today's technical challenges lie in how to develop low-power, high-accuracy cognitive computing, including new computing architecture circuit designs and algorithms. In the future, artificial intelligence chips will use specific algorithm accelerators to accelerate various neural network algorithms including Convolution Neutral Network and Recursive Neutral Network. The biggest advantage of the dedicated chip lies in its lower cost and power consumption, which will greatly increase the efficiency of the artificial intelligence algorithm.

Although the current mainstream research in the field of artificial intelligence is artificial intelligence operations on servers, there are more and more applications that must perform real-time operations on the terminals. This kind of computing architecture is called edge operation. This development trend will change the design and technical requirements of the overall artificial intelligence computing system architecture.

In addition, with the introduction of artificial intelligence technologies, IoT terminal devices have also been upgraded to various AIoT intelligent machines. Among them, service robots will be the first to become the new battlefield of AIoT. IFR estimates that the value of global service robots will be approximately 19.6 billion U.S. dollars from 2015 to 2018, while home service robots will account for 74% and entertainment and leisure robots will account for 26%. 2017 will be a year of contending for these two robots. Home service robots such as Amazon Echo and Asus COCOROBO, as well as American Jibo and France’s Buddy, will play a smart family hub.

(Original title: AIoT is gradually upgraded by artificial intelligence chips to help the Internet of Things.)