Why do we need artificial intelligence in smartphones and what it can do

The development of artificial intelligence systems has become a trend in the past few years. This can be seen especially clearly on the example of smartphones – AI is used to optimize many tasks. Let’s see where exactly it is applied and what prospects await us in the future.

How does AI work and where is it used?
To understand what artificial intelligence is in a smartphone, you first need to understand the concept of a neural network . In essence, it is a simplified model of the human brain. Our brains are made up of 90 billion neurons , which are cells that process and transmit electrical signals. They communicate with each other using the so-called synapses. Only in the case of a neural network, neurons are special cells that can be assigned numerical values.

The simplest model of a neural network (source – deep-review.com)

In a real brain, the same happens – an electrical signal, following from one neuron to another, will always be converted in the same way. That is why it can be difficult for us to break a persistent habit or adapt to a non-standard situation. For example, reading this article, you already start to get bored: the signal gradually fades away, but other impulses (watch a funny video with cats), on the contrary, intensify. If you are still interested in how neural networks are trained, we recommend reading specialized material . And we will move on to practical implementation.

OmniStudio-Developer PDF Dumps

PT0-002 PDF Dumps

SC-900 PDF Dumps

CIS-Discovery Exam Dumps

NSE7_PBC-6.4 Exam Dumps

PMI PfMP Exam Dumps

NS0-592 Practice Test

PCSAE Practice Test Questions

CIS-PPM Practice Questions

CIS-APM Test Questions

77200X Test Questions

H35-561 PDF Questions

NSE7_EFW-6.4 Questions

AZ-140 Questions PDF

PL-600 Questions

For object recognition, so-called convolutional neural networks are now used . It is with the help of them that the smartphone understands what is shown in the picture. The camera application in real time identifies an object or scene (animal, person, landscape) and sets the optimal settings – optimizes color, brightness and contrast, exposure.

Huawei camera software prompts for text on screen and optimizes the image

The same principle is used for face unlocking in most smartphones. The front camera takes a picture, and algorithms compare it with a given image using key points. A neural network always gives a probability, not an exact result: if it is large, the phone will be unlocked. Machine learning algorithms help if you’ve put on glasses or grown a beard. Without recognizing you, the system will prompt you to enter the password manually, and then the AI ​​makes adjustments to the model, adapting to the changes in appearance.

6.4 “HUAWEI P40 Lite 128GB Smartphone Green 18 999 *
In smartphones of the expensive segment, a different method is used. For example, in the iPhone 12, a special True Depth camera projects tens of thousands of points onto the face and builds a three-dimensional model. When trying to unlock, the system compares the face model with the original one. This method is more reliable – a fraudster will not be able to deceive the system by bringing a photo of a face to the camera.

Speech recognition works on the same principle. The only difference is that Google Assistant and Apple’s Siri send requests to cloud servers, where they are processed much faster than on your phone.

Neural networks of voice assistants are trained using huge amounts of information: from classical literature to live texts from the Internet.

Learning technologies are also used by machine translators. For example, Google Translate uses recurrent bidirectional neural networks to translate entire sentences using context. Previously, this happened word by word, so the quality of translation suffered in complex semantic constructions.

Instant translation using a camera is also available, but this requires the Internet, since the data is also processed on a remote server.

AI algorithms in modern smartphones are used for a variety of tasks. In addition to the ones we mentioned, they are used, for example, to track the movement of objects in real time (recognition of gestures and emotions, help with focusing videos), adjusting sound to the environment, improving augmented reality filters in AR-enabled applications, in systems navigation (to get directions without traffic jams) and many other cases.

Iron component
Artificial intelligence is also being used to optimize the smartphone itself. A processor typically uses 6-8 cores, two of which are high-performance and the rest are energy efficient and operate at a lower frequency.

The Snapdragon 730 processor has two efficient 2.2 GHz cores and six energy efficient cores running at 1.8 GHz

The neural network analyzes the user’s habits to ensure the optimal balance between battery consumption and performance. The smartphone detects which applications are used more often and at what intervals, then they are automatically loaded into RAM.

When listening to music or reading web pages, a lot of resources are not required, so only a couple of low-frequency cores are involved. And for games and resource-intensive programs, productive cores are periodically turned on.

To speed up the processing of computations for artificial intelligence, smartphone manufacturers began to allocate a separate computing unit for these tasks. The first processor with NPU (Neural Unit) was Huawei’s Kirin 970, released in late 2017. This chip parallelizes a huge number of small operations that are performed simultaneously. The central processor is not suitable for this – it has only 8 cores. The graphics accelerator contains thousands of cores, but consumes too much power.

Then other giants of the industry moved in. Apple used the A11 Bionic processor in the iPhone X, integrating the Neural Engine into it, which is capable of up to 600 billion operations per second .

A11 Bionic Chip with Neural Computing Accelerator (NPU)

Qualcomm has implemented hardware support for machine learning algorithms in processors starting with the Snapdragon 660.

Google has built a special Pixel Visual Core chip into its smartphones, which speeds up photo processing.

Pixel Visual Core

Thanks to him, HDR + images in the proprietary camera application are processed 5 times faster than using a regular CPU. The phone takes up to 16 photos with different exposures in a short period of time, and then combines them using a neural network.

Currently, the fastest processor in the world is Snapdragon 888 . At his presentation, much attention was paid to the capabilities of the new Hexagon 780 neural accelerator. Qualcomm claims that its performance is so high that the AI ​​”in real time can erase a specific person from a video or insert someone else.”

Prospects for the future
Progress is moving towards the fact that smartphones will soon be able to run even deep machine learning (the so-called Deep Learning ). Simply put, the number of layers of neurons will increase – networks will be able to perform more complex tasks.

For example, the front camera will constantly analyze the wearer’s face in order to understand his physical condition. Speech recognition accuracy will be improved, while the NPU will better understand the specific intent of the user. The proliferation of 5G networks will allow faster interaction with cloud servers.