By Dr. Ryan Chen, General Manager of the Computing and Artificial Intelligence Technology Group at MediaTek
In layman’s terms, “edge” processing is “local, in-device analysis.” This means that data is sampled and handled locally without the need to upload data beyond the device to cloud services for further processing.
Hybrid Cloud computing is optimally decentralized where a percentage of the data is still processed locally. This saves on bandwidth between the edge device and Cloud service, as well as acts to hide latency where possible. It also reduces the issue of packet loss during data transmission and enhances data security.
Edge AI is being increasingly heavily favored for the Internet of Things, in smart home devices, and is a necessity for autonomous vehicles. However, with the continuing improvement of edge AI processing capabilities and the growing spread of applications, new use-cases and services have been growing in the smartphone industry.
Have you read “MediaTek launches newest gaming chipsets, Helio G70, G80”?
Purchasing direct from a photo
The MediaTek Helio G90T chip currently features the most powerful edge AI hardware available today in a smartphone. On application it supports is real-time object recognition. If your photo includes a product, the system can automatically identify and analyze the brand, color, and size of the product relative to the person in the shot.
Furthermore, it can even provide the style, price, and even consumer product review results. This greatly shortcuts the “that looks great” to purchase. This is one practical application of edge AI that greatly improves the user experience. Data shows that for terminal devices with Edge AI-enabled, edge processing performance provides 200 images per second performance and accuracy has increased from 50% to over 99%.
While the initial image may be streamed from an internet service, the smart recognition factor is entirely processed on-device. This improves the overall user experience: low latency and fully reactive to what the user wants.
Turn on the TV simply by looking at it
MediaTek is providing the power for Edge AI applications in the smart home and car too. For example, the latest MediaTek MT8175 is an AI vision platform that can be used in smart TVs, smart cameras, and in-vehicle entertainment systems, etc., thus bringing the AI experience from smartphones to other smart-screen devices.
Soon Smart TVs might do away with the traditional remote control, instead opting for facial recognition and voice control. You can control the TV by asking it to change the channel or by gestures such as waving left-and-right to switch channels or up-and-down to adjust the volume.
In addition, real-time intelligent scene detection can customize the picture quality parameters to match the content you’re watching. For example, the TV will turn on only if mum or dad decides it – not the kids, your favorite show will play during dinner, news during breakfast morning, — there are many customization potentials to make it more in-tune with the home it resides in.
Next-generation edge AI in Smart Speakers
MediaTek edge AI already features in its smart voice assistants, but its capability is going to increase exponentially in upcoming devices. For example, MediaTek MT8518 voice AI SoC that debuted during CES 2019, features powerful edge-AI capabilities and more advanced features designed purely for voice-assistant devices. It supports an ultra-low power voice wake-up from standby, new low-power far-field commands, locally processed voiceprint recognition, and voice commands, etc.
Simply speaking, upcoming voice-enabled products using the MT8518 (smart speakers, smart translators, voice-enabled watches) will be able to distinguish the voices between owner and guests, automatically cancel-out background noises, provide voice condition recognition, and a degree of customization rather than being purely reliant on the Cloud-service. MediaTek is building this highly-capable new platform based on its existing global market share leadership in voice assistant devices.
AIoT gives us an ‘All AI Future’
MediaTek is pushing IoT beyond simple devices, adding intelligence and processing power where it wasn’t before. This Application IoT (AIoT) blends Edge AI + Internet of Things. For example, there is no need to install a discrete voice assistant device in every room in the future, it could feature among the smart home products such as light-bulbs, the TV, bedside clocks, etc.
In a world of application of IoT devices, there is no light switch. Simply saying “turn off the lights” while in the room triggers a small, AI accelerator chip embedded in the light bulb itself to act. This chip will have the capability for 20-30 keywords that it can react to.
At present, edge computing has been able to smoothly transfer partial data-processing from the traditional cloud to the terminal device, thus improving cost and efficiency. However, if you combine big data and artificial intelligence into this process, it will greatly improve the computational efforts with direct gains in efficiency.
MediaTek Edge AI has already been applied outside the smartphone market, e.g. automotive electronics and smart homes, while the company has also publicly stated that, “We want to become a key enabler for edge-enabled AI in devices.”