Monday, October 23, 2017

What are Artificial Intelligence processors on a mobile


In recent months, we've heard a lot about using specialized silicon for automatic learning on mobile devices. The new iPhone has its "neural motor," Huawei's Mate 10 comes with a "neural processing unit," and companies that manufacture and design chips (such as Qualcomm and ARM) are gearing up to supply hardware optimized for artificial intelligence to the rest of the industry.
What is not clear is how much it all benefits the consumer. When you buy a phone, should there be an "artificial intelligence chip" on your wish list? No, but let's dig a little deeper.

Why do we need these chips?

The reason for having AI mobile chips in the first place is quite simple. The common CPUs found on phones, laptops and desktops are simply not suited to the demands of automated learning , and try to make them end up with a slow service and a battery that is downloaded fast.
Current artificial intelligence requires computers to do many small calculations very quickly, but CPUs only have a handful of cores available to do mathematical operations . This is why the industry loves graphics processing units or GPUs. They were originally designed to render video game graphics, which requires a lot of small calculations very quickly. Instead of a handful of cores, they have thousands.
Now, installing thousands of cores on a chip for your phone is something that will not happen. But there are other architectural changes you can make to increase the amount of simultaneous work your chip can do. Qualcomm's head of computer and automated learning, Gary Brotman explains, " I think paralleling is certainly key, and doing it efficiently, especially ." However, he quickly adds that dedicated AI units are not the only way forward, as other fragments of chip architecture can also be adapted.
"AI Chip" is a useful term, but it is also imprecise. In the case of Huawei and Apple, what is offered is not a single , standalone chip , but dedicated processors that come as part of a larger SoC (or chip system) such as the Apple Bionic A11 . SoCs already contain several specialized components for things like rendering graphics and processing images, so adding a few cores for AI is somewhat similar to the course.

What do we get out of this?

As mentioned above, specialized AI hardware means, in theory, better performance and better battery life . But there are also advantages to user privacy and security, and to developers as well.

Artificial Intelligence = Greater Performance and Privacy

First, privacy and security. At the moment, many automated learning services must send their data to the cloud to perform the actual analysis. Companies like Google and Apple have devised methods to perform such calculations directly on the phone, but are not yet widely used. Having dedicated hardware encourages more AI in the device, which means less risk to users of data being filtered or pirated .
Also, if you do not send data to the cloud every few seconds, it means that users can access offline services and save data . That last part is a great help for developers as well. After all, if the analysis is done on the device, it will save the people running the application from paying the servers. As long as the hardware is up to speed, everyone benefits.

Is it ready to use?

The next section is where things get more complicated. The fact that a phone has an Artificial Intelligence chip does not mean that applications and services powered by Artificial Intelligence can take advantage of it .
In the case of Huawei and Apple, for example, both companies have their own APIs that developers must use to harness the power of their respective "neural" hardware. And before they can integrate that API, they should make sure that the artificial intelligence framework they used (for example, Google TensorFlow or Facebook Caffe2) is also compatible . If not, you will have to convert it, which also takes time.
"It will be some time before people develop experiences using this hardware. Until then there will be special partnerships between manufacturers and third parties, "says Anthony Mullen. That's why Microsoft is working with Huawei to make sure that its Translator application works offline with the company's NPU chip , and why Facebook partnered with Qualcomm to integrate the company's IA hardware.
Facebook worked with Qualcomm to make its self-augmented reality filters faster on the company's hardware. But while large companies like these can afford to invest in this , it is unclear if it will be worth the effort for every small application developer.
This will not be a problem for Apple , whose developers will only have to adapt their application once they use the company's Core ML framework; but could be a headache for Android, especially if different manufacturers start to introduce their own protocols.
Fortunately, Google is using its power over the ecosystem to combat this problem. Its mobile artificial intelligence framework, TensorFlow Lite, is already standardizing some experiences on mobile devices , and is introducing its own Android APIs to " leverage specific silicon accelerators ."
From a developer's point of view in the Android environment , it does not mitigate all the risks of fragmentation, " says Brotman. But it will certainly provide a construction to make it easier ." He adds that some of the effects of this work will not feel at all until Android P is ready.

Do I need this chip on my phone?

No, not really . So much work is being done to make IA services work better on the hardware currently available, unless you're a real advanced user, you do not have to worry about it .
In both Huawei and Apple cases , the main use of their shiny new hardware makes their phones better. For Huawei that means monitoring how to use Mate 10 over its useful life and reallocating resources to prevent it from slowing down; for Apple it means to have new features like Face ID or animojis.
Having computing power dedicated to the tasks of artificial intelligence is clear, but so are other features of high-end phones, such as double-chamber lenses or waterproofing. Skipping AI chips makes it good marketing now, it will not be long before it becomes another component .
Copyright © 2015 NewCydiaTweaks | Privacy Policy Privacy Policy | Disclaimer NewCydiaTweaks
Scroll To Top