Contact Us




Contactus:Jon Qiu


Contactus:James Mo



Add:B Building, Laifu Industrial Park, Liyuanxilu, Heping, Fuyong, Baoan, Shenzhen, China.

Home > Knowledge > Content
How is the AI of mobile phones popular? Jul 05, 2018

In 2018, a smartphone can have no full screen, but there must be no AI. But for most consumers, AI is a mysterious and mysterious existence. What exactly is AI? What is the role of AI? Can AI at this stage really be called AI? This series of problems has always plagued consumers.

The cause of consumer confusion may not be found. On the one hand, AI is a non-figurative existence. We can see the increase in screen ratio brought by the full screen at a glance, but it is difficult to specifically perceive the obvious role of AI. Perhaps for many consumers, AI is still equivalent to voice assistant. The so-called AI beauty, AI photo, compared to the improvement brought by the past Mito algorithm, may not be obvious.

On the other hand, the AI market is mixed. After all, AI is a hot spot. From the perspective of marketing, no AI is not easy to sell, and AI is bound to be moved into the promotion of various products. For consumers, I have been used to the experience of AI propaganda. The sense of difference in this is also the key to consumers' distrust of AI.

But AI technology is really real, and it is changing the experience of all aspects of the smart machine. In a sense, the smart machine itself is the carrier of AI representation. From camera experience to system optimization, from battery life to SoC computing, AI is everywhere. After throwing away the products on the market for AI and AI, we can better see the meaning of AI for smart phones.

Why is AI running on the phone?

AI, or artificial intelligence, is a neural network model and an algorithm.

The first generation of smartphones with voice assistants represented by the iPhone 4S, its AI application (voice assistant) is running in the cloud. Limited by the lack of network speed and semantic understanding, the experience is not good. Today's AI on smartphones writes neural network models into smartphones. Once the system detects that an AI application is needed, a computational analysis is performed.

At this stage, the reason why AI runs on the mobile phone side is that the AI operation with image recognition as the main application puts extremely high demands on the calculation speed. If you move the computing process to the cloud, there will be a delay, and the experience is unacceptable. Second, to maximize the ability of AI, it is necessary to have private data from consumers. For the protection of data, AI operations can only be selected to run on the mobile phone.

Running the AI application on the mobile phone does not solve a problem, and the energy efficiency ratio (which can be simply understood as the operation efficiency) is reasonable. Traditional CPUs and GPUs can perform AI calculations, but after all, they are not specifically designed for computing AI. Hard-running AI applications not only consume a lot of power, but also cause operational delays. How to make AI operations more efficient has become an urgent problem to be solved.

Optimizing the SoC is a better solution simply by improving the algorithm and not meeting the needs of the manufacturer. However, mainstream chip vendors have embarked on a different path in how to optimize SoCs.

The SoC represented by Huawei Kirin 970 and Apple A11 Bionic chose a method to add a dedicated AI chip. Kirin 970 has built-in artificial intelligence NPU module, and Apple A11 Bionic has built-in neural network dedicated acceleration module. Only the Kirin 970NPU chose the open NPU port, and the Apple A11 Bionic's neural network-specific acceleration module is currently only available for Face ID.

Adding a dedicated AI computing chip will undoubtedly bring about a significant increase in computing energy efficiency, solving the problem of efficiency and power consumption in one fell swoop. However, the introduction of a dedicated AI computing chip also brings about an increase in the cost of the chip.

SoCs represented by Qualcomm Snapdragon chips have chosen heterogeneous computing methods to handle AI applications. That is to say, through the Xiaolong artificial intelligence engine to judge, the most suitable kernel is selected from the CPU, GPU, and DSP for data processing to avoid waste of computing resources.

The third type of program represented by MediaTek P60 can be seen as a compromise between the above two options. MediaTek P60 introduced the APU design. However, the APU seems to be an independent AI computing chip. In fact, it is still an improved version of the DSP. Although it has a certain AI computing power, it is not as efficient as the NPU.

In order to quantify the performance of the current intelligent machine AI, such as Lu Master introduced the AI mark 1.0 running sub-system. Of course, this running sub-system is only analyzed from the aspect of AI computing efficiency. Considering the actual situation of AI application, the significance of AI running points at this stage is not great.

AI application scenario

Most of the AI in the smartphone is used for image recognition. Specifically, it is used for taking pictures. AI photography has become a must-have feature of today's smartphones. Up to the high-end flagship, down to thousands of dollars, no exception, will be carried.

Take scene recognition technology as an example. How to make consumers take a good photo in the easiest way is always a big problem that plagues manufacturers. Photographing scenes and environments are ever-changing, and a single camera model clearly cannot solve the problem. To this end, the solution given by the manufacturer is modeled according to different scenarios and different environments. When the AI program detects different scenes, the corresponding model can be called up as needed.

The same is true for portraits. The AI algorithm identifies the boundaries between people and the environment, and achieves accurate ambiguity. AI can detect the gender, age and facial features to give the best beauty plan.

In the past two years, the improvement of the overall camera capability of the smart phone has not only benefited from the improvement of the hardware level, but also the blessing of the AI algorithm. For consumers, the cost of taking good photos in the future will be lower and lower. AI will realize the dream of the consumer's photo master.

But is the role of AI limited to taking pictures? Nature is not. Taking Huawei Mate 10 as an example, by calling the NPU, the processing time is significantly shortened in scenarios such as offline translation, QR code scanning, and software startup speed. At present, the advantages of NPU are more obvious.

The future of AI and smart phones

We must admit that the current AI application is still in its infancy. Whether it is an AI chip or a software algorithm, there is a big breakthrough. How the future AI will change the world, we can not accurately predict. But how AI will change the smart machine, but it can be seen one or two.

With the popularity of 5G technology, cloud computing will once again become a hot spot. After solving the privacy security problem, the AI operation in the future smart machine will be transferred back to the cloud. Large computers will be able to handle more and more complex AI applications. Once the power is guaranteed, the AI will open a new door for the smartphone. AI applications will become "simulation" applications, enabling smartphones to use software to perfectly mimic physical reality. In terms of taking pictures, the smart machine can eliminate the SLR. In terms of information processing, voice assistants can eliminate labor.

Further, AI will have real wisdom like Jarvis, the intelligent system in Iron Man. At that time, the intelligent opportunity became a personal assistant in the true sense, a unique existence.

The competition about AI has only just begun. For manufacturers, if they want to win in the AI competition, they must not only improve their R&D capabilities, but also integrate the supply chain, from chips to algorithms.

What is certain is that the AI has a future. It is able to integrate AI into the product perfectly, and fully build the AI ecosystem. The manufacturers that truly make consumers feel the convenience of AI can gain an advantage in the future competition. Today, Huawei and Apple have taken the lead, how will the rest of the manufacturers catch up, and what changes will happen in the future AI industry, which is worth looking forward to.