Saturday, November 16, 2019

𝟱𝗚 𝗡𝗲𝘁𝘄𝗼𝗿𝗸 𝗖𝘂𝗿𝗿𝗲𝗻𝘁 𝗦𝘁𝗮𝘁𝘂𝘀 𝗮𝗻𝗱 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 𝗣𝗿𝗼𝘀𝗽𝗲𝗰𝘁𝘀 𝗳𝗼𝗿 𝟮𝟬𝟮𝟬 | 𝗙𝘂𝘁𝘂𝗿𝗲 𝗶𝘀 𝟲𝗚

Interview with Anton Steputin, author of the book “Mobile Communications on the Way to 6G ”

Many equipment suppliers and mobile operators have taken the vector to the next generation mobile network. This is facilitated by the active work of standardizing bodies that have moved into the second (final) phase of the specification of 5G networks . Operators have conducted numerous testing of innovative solutions around the world. Currently, the first commercial 5 G swallows are appearing . What will this give subscribers? Than 5G networks different from previous generations? Are there any obstacles to the evolution of mobile communications? We’ll talk about this and much more with Anton Steputin, a member of the League of Technical Trainers, two-time winner of the “Trainer of the Year” competition in the nomination “Technical Trainer”, chairman of the Organizing Committee of the annual TELECOMTREND International Congress, candidate of technical sciences, associate professor SPbGUT them. prof. M.A. Bonch-Bruevich, the author of the book "Mobile Communications on the Road to 6G."



5G network updates



Let's start with the most important thing: 


What benefits will 5G networks give people?


The main services that require the creation of a new generation of mobile communication networks are as follows:


  1. ultra-wideband mobile communication (enhanced Mobile Broadband, eMBB),
  2. ultra-reliable Low Latency Communication (URLLC),
  3. mass machine communication (Massive Machine-Type Communications, mMTC).



If you operate with specific numbers, then they are as follows:

- low latency to ensure the transfer of critical information from critical IoT sensors :  up to 0.5 ms (for services of ultra-reliable machine communications - URLLC) and up to 4 ms (for services of ultra-wideband mobile communication - eMBB) ;

- massive IoT (mass connection of devices of the Internet of things). The total number of connected or accessible subscriber terminals per unit area (≥1 million / km 2 );

- high speeds: up to 20 Gbit / s downlink (ie, to the subscriber) and up to 10 Gbit / s uplink (in the opposite direction);

- Support for the mobility of a subscriber moving at a speed of up to 500 km / h.


Is there a need now for such speeds?


Let's look at the services that enter our lives and become more and more popular:


  • 360 degree video;
  • virtual and augmented reality;
  • 4K and 8K video;
  • tactile internet;
  • smart and safe cities and enterprises;
  • 5G will be in demand on vehicles that are increasingly equipped with telemetry systems, and in the future should become unmanned;
  • 5G will also contribute to the emergence of new services, such as holographic calls.

In addition, the drivers of growth in mobile traffic consumption are:

  • popularization of cloud technologies - models of online storage of subscriber data on numerous servers distributed on the Internet;
  • online games and their updates;
  • increase in the number of devices;
  • an increase in the consumption of video services and an increase in the resolution of the video image, which was mentioned earlier.


The main thing is to provide the subscriber with a channel. As practice shows, no matter how wide the channel is, it will occupy it. But first of all, 5G will be in demand for the B2B segment.

How will such peak speeds in 5G be achieved: 10-20 Gbit / s?


1. Massive MIMO, as well as adaptive methods of beamforming and tracking (beamforming).
These solutions are necessary to ensure reliable and sustainable mobile broadband in the higher frequency ranges. The Massive MIMO class includes a system with the number of controllable antennas> 8. Moreover, each “controllable antenna” can be a group of radiating elements.

2. New Radio (NR).
Unlike the LTE radio interface, where the LTE subcarriers are almost always separated by an interval of 15 kHz, the distance between the subcarrier frequencies can vary in 5G NR. So 5G NR subcarriers can be spaced 15 kHz x 2n apart. The maximum distance between subcarriers in 5G NR is 240 kHz. This will allow you to easily configure the network behavior when providing various types of services that have different delay requirements. In addition, in Release 16 it is planned to consider alternative options for the radio interface for 5G networks, which will improve the efficiency of using the precious frequency spectrum.

3. A wider frequency spectrum to be allocated in the high frequency ranges.

You correctly noted that these are the maximum speeds that are targeted for the 5G standard. The final user speed that the subscriber will receive is influenced by many factors:


  • The number of subscribers in the base station sector;
  • Category of equipment used by the subscriber;
  • Remoteness from the serving base station;
  • Terrain and / or walls and even windows of buildings;


What innovative technologies besides those described above will appear in 5G?


1. Network slicing

This means that the 5G network infrastructure can be logically cut into “network layers” - “slices” for different business applications and for different radio access technologies. 

These networks can be separately optimized for various data rate requirements. So for watching a 4K video, speed will be important, but delays due to the inertia of the perception of the visual apparatus will not be critical. To transmit data critical to delays, a special slice called ultra-reliable low latency communication will be used. These types of data include tactile internet, online games, etc.

2. D2D (Device-to-Device)

This is a direct device-to-device interaction without a base station. 

D2D, in particular, is necessary for the interaction of an unmanned vehicle with road infrastructure and other vehicles. In addition, D2D technology will be used in the framework of the concept of public safety (public safety) in the direct interaction of two subscriber terminals for special purposes when deployed in an emergency.

The following solutions will complement the above technologies:

3. Network Functions Virtualization (NFV)

It involves the use of virtualization technologies to separate the functions of logical network elements from the hardware infrastructure of a communication network. When using NFV in telecommunication networks, network functions are performed using specialized software models that run on servers or virtual machines in telecommunication networks.

4. "Fog computing" (foggy computing).

The deployment of micro-data centers is as close as possible to the place of traffic generation (Mobile Edge Computing), in particular, directly at the base stations.


How is the 5th generation network specification going?



In December 2017, 3GPP agreed on the first version of the 5G Release 15 standard for the non-standalone use case (i.e., together with the LTE network ). Several scenarios of such interaction are specified.

In June 2018, 3GPP released the second version of Release 15, already for stand-alone use (standalone mode). 

The standalone architecture of 5G networks is the architecture within which the base stations are directly connected to the core of a new generation network (NGC) via the control plane (NG-C) and user plane (NG-U) interfaces.

And in July 2019, 3GPP promises to release Release 16, which will include all aspects of 5G: not only advanced mobile broadband access services (eMBB), but also highly reliable low-latency communications (URLLC) and mass inter-machine communications (mMTC).

What hinders the development of 5G networks? 


I will name three main factors that impede the more rapid development of fifth-generation networks:

  • Uncertainty with frequency ranges and, accordingly, the lack of network infrastructure and subscriber equipment;
  • Lack of a regulatory framework governing the operation of 5G networks;
  • 5G launch and the most effective commercialization should be based on the most relevant for subscribers and businesses options for using new networks (use cases). The sooner they are identified and identified as a technology strategy, the faster 5G networks will evolve.

 Should we expect the creation of telecommunications equipment for 5G networks by our country? 


First you need to decide what is the domestic industry and domestic telecommunications equipment, in particular? Regulatory acts are necessary in which the criteria for classifying manufactured products as domestic will be clearly spelled out. Otherwise, all world manufacturers with one or another level of localization in our country will fall there.

Nevertheless, there are originally telecommunication companies, and they cover part of the needs of operators in certain areas. In particular, equipment and software (software) of the transport network, data processing centers (servers, data storage systems), virtualization software (NFV / SDN), etc.

The launch and development of the 5 G network is an expensive , in particular, in connection with the use of higher frequency ranges ...


Yes, base stations will need a lot.

Various approaches are possible to reduce the cost of deploying 5G networks.


  • Build part of the infrastructure at the expense of the infrastructure operator. In particular, it makes no sense to lay cable in the subway to all operators. The flip side of launching the entire infrastructure on the basis of a single infrastructure operator is the lack of competition in the radio coverage of the network. All operators will provide the same 5G signal level throughout the presence of the infrastructure operator. As part of this solution, mobile operators will be able to compete with each other only with the services provided. There will also be problems in terms of aggregation of the frequency spectrum that the operators already have with the spectrum that they will receive for the development of 5G networks. Therefore, in my opinion, the infrastructure operator is appropriate for deploying about 40-50% of the network infrastructure and mainly in the low-frequency ranges, where it is not possible to allocate to all operators the necessary frequency bands for 5G. Moreover, this issue should be addressed immediately at several levels: specialized departments and organizations, shareholders of mobile operators together with specialized technical specialists based on the technological development strategy that the operator will determine for himself.
  • Provide special places for the placement of communication equipment in the construction of new buildings.
  • To build equal access for all operators to the intra-house infrastructure.




Where in the first place will 5G networks be launched : in cities, suburbs, rural areas or at critical facilities (gas pipelines, oil pipelines, etc.)?


Definitely, in the first place, 5G networks will be launched in large cities. This will give value (profit) and image for mobile operators. However, it is highly undesirable to widen the already significant digital divide between rural and urban areas. Access to information enables people to grow.

With regard to specific numbers and terms, according to the state program Digital Economy of the Russian Federation approved by the government, in 2020 5G networks should work in 8 cities of Russia, and by 2025 in all cities with a population of more than 1 million people.

Thus, with the advent of fifth-generation networks, traffic consumption, according to some analytical companies, will increase 20 times by 2025.

By 2023, according to Ericsson's forecast, 1 billion connections will be registered in 5G networks. And services based on 5G by this time will be available to 20% of the global population.

One of the most valuable that an operator has is a frequency resource. What should it be to run 5G networks? Is there a magic source of new frequencies for 5G networks?


In 2019, the next World Radiocommunication Conference (WRC-19), which takes place every four years, will be held.The most likely for harmonization are the radio frequency bands: 24.25-27.5 GHz, 37-43.5 GHz.  

The 70/80 GHz bands can also be harmonized at WRC-19.

Harmonization of the 31.8-33.4 GHz band for 5G at WRC-19 has practically no chance, since ensuring compatibility with existing services is problematic.

As for the lower frequencies, for 5G there is also no final solution for the frequency ranges and bandwidth. So, in Russia, most of the 3.4-3.8 GHz band is occupied by the military.

The LTE Union has proposed the use of a frequency range of 4.4-4.99 GHz for the 5G, which is not standard for the European Region. This will increase the frequency spectrum for the deployment of new generation mobile networks. However, there is no commercial equipment for this frequency range.

It is expected that based on the results of WRC-19, the State Commission on Radio Frequencies (SCRC) will make a decision on frequencies taking into account the 5G development concept, which is being prepared by the Ministry of Digital Development, Telecommunications and Mass Communications within the framework of the national program “Digital Economy of the Russian Federation”. It is expected that in the upper frequency ranges the distribution of frequencies between all interested operators will be implemented on the basis of tenders.

Operators with the introduction of 5G networks will also be forced to cope with the management of a whole "zoo" of technologies: 2G, 3G, 4G, 5G. Maybe it makes sense to abandon outdated and unpromising technologies? And which of the technologies is the first to drop out: 2G or 3G.


The first candidate to drop out I see 3G . 2G technology in the 900 MHz band will live long due to the fact that it has a fragmented spectrum and subscribers work there and will continue to work for some time: old terminals and M2M sensors.


Currently, the question of the prospects of Wi-Fi technology in anticipation of the launch of 5G mobile networks is increasingly being addressed. Is Wi-Fi technology coming to an end when 5G networks provide people with high-speed Internet with low latency? 

Already now, in most cases, LTE networks provide high speed performance. Some people already stop using Wi-Fi at home. But these are far from all people. 

In order for Wi-Fi to “die”, giving way exclusively to mobile networks, it is necessary for several events to happen:


mobile networks everywhere must provide the same quality of customer service;
mobile operators must provide tariffs that will allow subscribers to pump out the necessary amount of traffic, similar to the volumes pumped through Wi-Fi networks.
Thus, each technology has its own niche. One (Wi-Fi) - provides local coverage, guaranteeing certain speed indicators to the subscriber. Another - provides global coverage at almost anywhere in the world map, wherever we are.

Moreover, technology can complement each other. In particular, 3GPP specified such solutions as Wi-Fi Offloading, LWA (LTE-WLAN aggregation), LAA (Licensed Assisted Access) and eLAA (Enhanced Licensed Assisted Access). These technological solutions allow you to use either Wi-Fi access points to offload the radio interface of mobile operators, or use the unlicensed Wi-Fi frequency range for LTE networks.


A number of equipment suppliers have also announced a solution such as MuLTEfire , which allows LTE networks to be deployed exclusively in the unlicensed frequency range.

Friday, November 8, 2019

𝟮𝟬𝟮𝟬 𝗧𝗲𝗰𝗵𝗥𝗲𝘃𝗶𝗲𝘄 𝗨𝗽𝗱𝗮𝘁𝗲𝘀 𝗼𝗻 𝟱𝗚 𝘁𝗼 𝗣𝗼𝘄𝗲𝗿 𝗧𝗵𝗲 𝗜𝗻𝘁𝗲𝗿𝗻𝗲𝘁 𝗢𝗳 𝗧𝗵𝗶𝗻𝗴𝘀 𝗜𝗢𝗧 | 𝗙𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝟱𝗚


In 2022, 550 million devices of the Internet of things will be connected to 5G networks

(Ericsson forecast)


10 years - just so much goes to the telecommunications industry to introduce a new generation of communication since the 1980s. According to this schedule, the 2020s promise to become a decade of the 5G (5th Generation) standard - networks with a bandwidth of up to 20 Gbit / s.



Its implementation will not only accelerate the mobile Internet and the quality of communication on our smartphones - 5G is intended to become an infrastructure for key technologies of the future: virtual and augmented reality , unmanned vehicles, Internet of things , etc., followed from explanations that the International Telecommunication Union (ITU) 2015.

The technological development of the standard should be completed by 2020. The data transfer rate will increase by 30-50 times in comparison with the previous generation. If the 3G standard reduces the signal delay to 100 milliseconds, 4G to 10 milliseconds, then the 5G is only 1 millisecond.

5G IOT
IoT and 5G


5G  4G ALLOWS DOWNLOAD THE TWO-HOUR FILM IN HIGH-PERMISSIBLE FOR 6 MINUTES, 

AND 5G - EVERYTHING FOR 15 SECONDS


The new standard will have to work with a large amount of data: according to Cisco, in 2020 global mobile traffic will grow to 30.6 exabytes per month (1 exabyte - 1 million Tb) - this is 8.3 times more than in 2015 (3.7 exabytes). Also, from 7.9 billion to 11.6 billion, the number of mobile devices connected to the Internet will increase, not only smartphones and tablets, but also items of the Internet of things, for example, gadgets for the "smart" home. Ericsson expects that by 2022, 550 million of these devices will be connected to 5G.

A single network is necessary for the harmonious operation of the ecosystem of new gadgets, which today rely on different protocols. The same applies to unmanned vehicles: for each drone, every day it will be necessary to transfer and process terabytes of data (not only from the sensors of the car itself, but also from cameras, radar, etc.). According to the German automaker BMW, after 2020, the fifth generation network will link up to 70 million "autopilots". Finnish Nokia called 5G a chance to reduce the accident rate on the roads to zero. The technology should help in the development of agriculture (remote management of machinery, monitoring of farmland using drones), industry (management of robotic assembly production, industrial 3D printers), medicine (remote operation via the network in real modetime ) and the entertainment industry (computer games with virtual reality technology with a high level of graphics, high-definition video transmission without delay).



All key markets related to 5G, in the long-term 2020 will surely grow, analysts expect Gartner. Thus, the telecommunications industry will increase from $ 1.4 trillion in 2015 to $ 1.7 trillion in 2020 (the average annual growth rate is 3.9%). The market of wearable devices ("smart" glasses, watches, etc.) over these years will grow from $ 12.5 billion to $ 37.1 billion; the market of equipment for the construction of mobile networks - from $ 40 billion to $ 60 billion; the market of voice services - from $ 900 million to $ 1.1 billion; the mobile applications market - from $ 45.3 billion to $ 162.5 billion.

While 5G is the technology of the future ? Inventions 2019


 its application is already widely tested on a limited scale. The development of fifth-generation networks is carried out by all the leading telecommunications operators in the world - American Verizon and AT & T, British Vodafone, Scandinavian Telenor and Teliasonera, etc. The Chinese market for Huawei and ZTE, Korean Samsung, European Nokia and Ericsson and American Cisco and Qualcomm . Follow the development of technology and technology giants Silicon Valley. For example, Google in 2015 launched a secretive program for managing drones through a fifth-generation connection called SkyBender.

China has already established near Beijing, Huairou, the world's largest test station 5G: Ericsson, Huawei, Nokia, ZTE, DTT and Intel are participating in the network testing, Xinhua News Agency reported in March with reference to the Ministry of Industry and Information of the country.

Korean operator SK Telecom in November 2015 announced the achievement of a data transfer rate of 19.1 Gb / s. In 2017, the company hopes to launch the first commercial technology tests. At the same time, Verizon is counting on the commercialization of 5G. In 2018, the innovative communication promises to provide the participants and spectators of the Winter Olympics in Pyeongchang Korean operator KT.

MTS and Megafon announced similar plans for the 2018 World Cup in Russia (Megafon in June, and MTS in September 2016 already conducted the first tests of 5G, working on the development of technology and other participants of the Big Four - Vimpelcom and Tele2). The Tokyo Olympics in 2020 will be covered by the operator's 5G network of NTT DoCoMo. China Mobile in December 2016 announced the commercial launch of the standard in China in 2020.

$5 billion
IN YEAR MUST COMPOSE UP TO 2020 SCIENTIFIC RESEARCH CONSUMPTION AT 5G




All these plans will require large-scale investments: according to Markets Reports Hub, the consulting company, until 2020, research expenditures for 5G should be at least $ 5 billion per year.

Technology developers are left with many regulatory issues: 

under the auspices of ITU, governments and businesses will have to unify the 5G standards, allocate frequencies in higher bands (new base stations will need to be built), maintain and update the mobile infrastructure (5G networks will work in parallel with the already built networks of previous ones generations). Finally, developers have to believe that their hopes for a technological breakthrough of other new technologies that need the 5G properties will come true. This will entail a large-scale renewal of the mobile "iron" park: users who want to join the future through the fifth generation communication will have to acquire a new smartphone, laptop, fitness tracker , voice assistant, etc.

5g Security:


 As with other new technologies, the 5G carries an additional threat. 

 Intel analyst Matthew Rosenquist considers the most vulnerable industry for which the fifth generation connection increases the risks of data leaks, Internet of things (IoT). With increasing network speed, more physical objects will be connected to it. As a result, hackers will try to gain access to corporate and user information in such sectors as transportation (attacks on control systems of unmanned vehicles based on 5G), health (theft of health data of patients served remotely due to high-speed data transfer in new networks) and the transport of goods by drones (hacking of logistics systems and for the conduct of terrorist attacks using drones that quickly "seize" the signal in 5G networks), warned the expert.


Next we shall be covering 


Tuesday, November 5, 2019

The Present and Future Trends of Machine Learning on Devices [Update for 2020]

As you, of course, noticed, machine learning on devices is now developing more and more. Apple mentioned this about a hundred times during WWDC 2017. It's no surprise that developers want to add machine learning to their applications.APIs IoT big data and AI are driving forces for Machine Learning in 2020.

However, many of these learning models are used only to draw conclusions based on a limited set of knowledge. Despite the term "machine learning", no learning on the device occurs, knowledge is inside the model and does not improve with time.

Machine Learning Future Trends


The reason for this is that learning the model requires a lot of processing power, and mobile phones are not yet capable of it. It is much easier to train models offline on a server farm, and all improvements to the model include in the application update.


It is worth noting that training on the device makes sense for some applications, and I believe that with time, such training of models will become as familiar as using models for forecasting. In this text I want to explore the possibilities of this technology.

Machine Learning Future Trends
Machine Learning Future Trend

Machine Learning Today Deep Learning in Neural Networks an overview

The most common application of deep and machine learning in applications is now the computer vision for analyzing photos and videos. But machine learning is not limited to images, it is also used for audio, language, time sequences and other types of data. The modern phone has many different sensors and a fast connection to the Internet, which leads to a lot of data available for the models.

iOS uses several models of in-depth training on devices: face recognition in the photo, the phrase " Hello, Siri " and handwritten Chinese characters . But all these models do not learn anything from the user.

Almost all machine learning APIs (MPSCNN, TensorFlow Lite, Caffe2) can make predictions based on user data, but you can not get these models to learn new from these data.

Now the training takes place on a server with a large number of GPUs. This is a slow process that requires a lot of data. A convolutional neural network, for example, is trained on thousands or millions of images. Training such a network from scratch will take several days on a powerful server, a few weeks on the computer and for ages on the mobile device.

Learning on the server is a good strategy if the model is updated irregularly, and each user uses the same model. The application receives a model update every time you update the application in the App Store or periodically load new settings from the cloud.

Now training large models on the device is impossible, but it will not always be so. These models should not be large. And most importantly: one model for everyone may not be the best solution.


Why do I need training on the device? 

There are several advantages of learning on the device:


  • The application can learn from the data or behavior of the user.
  • The data will remain on the device.
  • Transferring any process to the device saves money.
  • The model will be trained and updated continuously.
  • This solution does not work for every situation, but there are applications for it. I think that its main advantage is the ability to customize the model to a specific user.



On iOS devices, this is already done by some applications:

The keyboard learns based on the texts that you type, and makes assumptions about the next word in the sentence. This model is trained specifically for you, not for other users. Since training takes place on the device, your messages are not sent to the cloud server .
The "Photos" application automatically organizes images into the "People" album. I'm not entirely sure how this works, but the program uses the Face Recognition API on the photo and places similar faces together. Perhaps this is simply uncontrolled clustering, but the learning should still occur, since the application allows you to correct its errors and is improved based on your feedback. Regardless of the type of algorithm, this application is a good example of customization of user experience based on their data.
Touch ID and Face ID learn based on your fingerprint or face. Face ID continues to learn over time, so if you grow a beard or start wearing glasses, it will still recognize your face.
Motion Detection. Apple Watch learns your habits, for example, changing the heartbeat during different activities. Again, I do not know how this works, but obviously training must occur.
Clarifai Mobile SDK allows users to create their own models for classifying images using photos of objects and their designations. Typically, the classification model requires thousands of images for training, but this SDK can learn only a few examples. The ability to create image classifiers from your own photos without being an expert in machine learning has many practical applications.
Some of these tasks are easier than others. Often "learning" is simply remembering the last action of the user. For many applications this is enough, and this does not require fancy machine learning algorithms.

The keyboard model is simple enough, and training can occur in real time. The "Photos" application learns more slowly and consumes a lot of energy, so training occurs when the device is on charge. Many practical applications of training on the device are between these two extremes.

Other existing examples include spam detection (your email client learns on the letters you define as spam), text correction (it learns your most common mistakes when typing and fixes them) and smart calendars, like Google Now , that study recognize your regular actions.AI and machine learning in 2018 2019 2020 are going to change alot.

How far can we go in Machine Learning ?


If the goal of learning on the device is to adapt the machine learning model to the needs or behavior of specific users, then what can we do about it?

Here's a funny example: a neural network turns the drawings into emoji. She asks you to draw some different shapes and learns the pattern to recognize them. This application is implemented on the Swift Playground, not the fastest platform. But even under such conditions, the neural network does not study for long - on the device it takes only a few seconds ( that's how this model works ).

If your model is not too complicated, like this two-layer neural network, you can already conduct training on the device.

Note: on iPhone X, developers have access to a 3D model of the user's face in low resolution. You can use this data to train a model that selects emoji or another action in the application based on the facial expressions of the users.

Here are a few other future opportunities:


  • Smart Reply is a model from Google that analyzes an incoming message or letter and offers a suitable answer. It is not yet trained on the device and recommends the same answers to all users, but (in theory) it can be trained on the user's texts, which will greatly improve the model.
  • Handwriting recognition, which will learn exactly on your handwriting. This is especially useful on the iPad Pro with Pencil. This is not a new feature, but if you have the same bad handwriting as mine, then the standard model will allow too many errors.
  • Recognition of speech, which will become more accurate and adjusted to your voice.
  • Tracking sleep / fitness applications. Before these applications will give you tips on how to improve your health, they need to know you. For security reasons, it's best to stay on the device.
  • Personalized models for dialogue. We still have to see the future of chat bots, but their advantage lies in the fact that the bot can adapt to you. When you talk to a chat-bot, your device will learn your speech and preferences and change the answers of the chat-bot to your personality and manner of communication (for example, Siri can learn to give fewer comments).
  • Improved advertising. No one likes advertising, but machine learning can make it less intrusive for users and more profitable for the advertiser. For example, an advertising SDK can learn how often you look and click on ads, and choose more suitable advertising for you. The application can train a local model that will only request advertisements that work for a particular user.
  • Recommendations are the widespread use of machine learning. The podcast player can be trained on the programs you listened to to give advice. Now applications are performing this operation in the cloud, but this can be done on the device.
  • For people with disabilities, applications can help navigate the space and better understand it. I do not understand this, but I can imagine that applications can help, for example, distinguish between different drugs using a camera.
  • These are just a few ideas. Since all people are different, machine learning models could adapt to our specific needs and desires. Training on the device allows you to create a unique model for a unique user.


Different scenarios for learning models


Before applying the model, you need to train it. Training should be continued to further improve the model.

There are several training options:

  • Lack of training on user data. Collect your own data or use publicly available data to create a single model. When you improve a model, you release an application update or simply load new settings into it. So do most of the existing applications with machine learning.
  • Centralized training. If your application or service already requires data from the user that is stored on your servers, and you have access to them, then you can conduct training based on this data on your server. User data can be used to train for a particular user or for all users. So do platforms like Facebook. This option raises questions related to privacy, security, scaling and many others. The question of privacy can be solved by the method of "selective privacy" of Apple, but it also has its consequences .
  • Collaborative training. This method moves training costs to the users themselves. Training takes place on the device, and each user teaches a small part of the model. Updates of the model are sent to other users, so that they can learn from your data, and you - on them. But this is still a single model, and all of them end up with the same parameters. The main advantage of such training is its decentralization . In theory, this is better for privacy, but, according to studies , this option may be worse.
  • Each user is trained in his own model. In this version, I am personally most interested. The model can be learned from scratch (as in the example with pictures and emoji) or it can be a trained model that is customized for your data. In any case, the model can be improved over time. For example, the keyboard starts with a model already taught in a specific language, but eventually learns to predict which sentence you want to write. The downside of this approach is that other users can not benefit from this. So this option works only for applications that use unique data.



How to Train on the Device to Learn?


It is worth remembering that training on user data is different from learning on a large amount of data. The initial model of the keyboard can be trained on a standard body of texts (for example, on all Wikipedia texts), but a text message or letter will be written in a language different from the typical Wikipedia article. And this style will differ from user to user. The model should provide for these types of variations.

The problem is also that our best methods of in-depth training are rather inefficient and rude. As I said, the training of the image classifier can take days or weeks. The learning process, stochastic gradient descent, passes through small stages. The data set can have a million images, each of which the neural network will scan about a hundred times.

Obviously, this method is not suitable for mobile devices. But often you do not need to train the model from scratch. Many people take an already trained model and then use transfer learning based on their data. But these small data sets still consist of thousands of images, and even so the learning is too slow.

With our current training methods, the adjustment of models on the device is still far away. But not all is lost. Simple models can already be trained on the device. Classical machine learning models such as logistic regression, decision trees, or naive Bayesian classifiers can be quickly trained, especially when using second-order optimization techniques such as L-BFGS or a conjugate gradient. Even the basic recurrent neural network should be available for implementation.

For the keyboard, the online learning method can work. You can conduct a training session after a certain number of words typed by the user. The same applies to models that use an accelerometer and traffic information, where the data comes in the form of a constant stream of numbers. Since these models are trained on a small part of the data, each update must occur quickly. Therefore, if your model is small and you do not have so much data, then training will take seconds. But if your model is larger or you have a lot of data, then you need to be creative. If the model studies people's faces in your photo gallery, it has too much data to process, and you need to find a balance between the speed and accuracy of the algorithm.

Here are a few more problems that you will encounter when learning on the device:


  • Large models. For deep learning networks, current learning methods are too slow and require too much data. Many studies are now devoted to learning models on a small amount of data (for example, in one photo) and for a small number of steps. I am sure that any progress will lead to the spread of learning on the device.
  • Multiple devices. You probably use more than one device. The problem of transferring data and models between the user's devices remains to be solved. For example, the application "Photos" in iOS 10 does not transmit information about people's faces between devices, so it learns on all devices separately.
  • Application updates. If your application includes a trained model that adapts to user behavior and data, what happens when you update the model with the application?

Training on the device is still at the beginning of its development, but it seems to me that this technology will inevitably become important in the creation of applications.


For more details you must check AI for Good. It is a United Nations platform, centered around annual Global Summits, that fosters the dialogue on the beneficial use of Artificial Intelligence, by developing concrete projects


Begins May 4, 2020
Ends May 8, 2020

Popular Posts