MTS Lead Architect Mikhail Mishurovsky on Collaboration with ITMO
About a year ago, the MTS company started focusing on the development of AI-based products. A research group was formed as part of this new direction. The group’s specialists create virtual assistants, incorporate AI technologies in medicine and office management, as well as carry out two joint projects with ITMO University. Mikhail Mishurovsky, the lead architect of MTS’s AI research group, gave an open lecture about the company’s key projects in the field of artificial intelligence, the collaboration with ITMO, and the specialists of the future. ITMO.NEWS publishes the highlights of his speech.
One of the biggest telecommunication companies in Russia, MTS is a digital company that conducts a range of projects in such fields as artificial intelligence, big data analysis, and cloud technologies.
To support such solutions and help them take over the market, the company has recently established the so-called Innovations Center. The Center’s specialists conduct projects in six key fields: cloud technologies, artificial intelligence, eHealth, education, eSports, and the MTS StartUp Hub accelerator.
Cloud technologies. What can be cloud-based technologies used for? There is a range of tasks that have to meet some conditions, for example, regarding information security. That’s where our services, which are both secure and efficient, may come in handy.
eНealth. This is a pretty new field, which has to do with such technologies as telemedicine, personalized medicine, and other related areas. Among the latest trends in this sphere are medical data processing solutions, medical analytics, and recommendation services. These are the technologies we work with as part of the eHealth track.
eLearning. You’ve probably heard about such a popular MOOC-platform as Coursera. We also develop electronic courses on various topics, from languages to modern technologies.
eSports. This track has to do with video games and cyberindustry.
MTS StartUp Hub Accelerator is a sort of a platform where young scientists and startups can share ideas and establish useful connections.
And last but not least, artificial intelligence. AI is a very broad term. There are many definitions to it, from traditional to sector-specific. We regard AI as a set of technologies, practices, and approaches that allow computational systems to interact with humans using natural languages, as well as ‘understand’ the world around them and adapt to it.
Why is this important? The thing is that we humans have emotions, and we express them using our language, and it’s crucial that robots understand it.
AI Center’s research group
MTS’s Innovations Center brings together developers, programmers, mathematicians, linguists, and products specialists. The main goal of the Center is to introduce modern technological solutions in the field.
We test all our solutions on the company’s services first. This allows us to enhance internal business processes, optimize our work, and increase productivity.
- Virtual assistant (chatbot, ASR, TTS, BSS)
Virtual assistants help people complete tasks in the digital world. A good example of such an assistant is Yandex’s chatbot Alice. Apart from common tasks, such as internet search, it can also run applications and chit-chat.
As of now, we only use our chatbot for internal needs. We’ve been working on it for several years and have managed to create a pretty powerful chatbot that is currently being used for communication with MTS clients. For example, if you want to activate Internet on your phone, or change your rate plan, or solve some other problem and get feedback, you most often have to deal with a chatbot. Its efficiency is estimated to be some 75%, which is a very good result for this kind of systems.
We also work with such technologies as speech synthesis and recognition, as well as blind separation of sources. Why so many aspects? Although there are many box solutions available, some of them have patents. When using such services, you have to meet certain conditions. For one, some of Google’s solutions use client-server architecture, which implies that the data is being sent abroad. Sometimes this can be an issue. That’s why, to work efficiently, we have to come up with our own speech recognition technologies.
- AI Medicine (Symptom checker, Medical CV)
One of our projects is called SmartMed, and it is aimed at solving various tasks in the field of telemedicine. Another project that we’re currently working on is called Symptom Checker. This is a decision making support system. It helps doctors assess the patient’s status and make a follow-up decision. This system doesn’t diagnose patients but can provide a second opinion.
And finally, medical computer vision. This is a technology that can be used for images recognition.
Document management (Legal Tech)
About half a year ago, our team started working in the Legal Tech sector. Legal technology, or Legal Tech, refers to the use of technology and software to provide legal services. We were surprised to learn that there are quite a lot of problems in the legal sphere that have to do with document management. AI technologies, in their turn, can make the process much easier for users.
All the projects are implemented based on technologies of speech recognition, natural language processing (NLP), computer vision and machine learning (neural networks, deep learning and other techniques).
In the sphere of ASR (automatic sound recognition), we want to achieve world-class speech recognition accuracy. There is a wealth of data now on this topic, so it’s easy for specialists in the field to compare, analyze, and implement a variety of classical approaches. The majority of this information is in English, though; there’s not a lot available in Russian: we don’t have any open databases, and finding valid information to compare the data with the baseline proves extremely difficult. There’s not a lot of few open trained models available, either. That’s why it’s so important to pursue development in this field.
We design technologies in a way to make them versatile for different sectors. We can use them as cloud solutions for the development of end devices. The requirements they’d need to meet are different in each case: for example, that the technology could be used offline, work slowly but with high precision, or, vice versa, work really fast but consume little energy, also being able to adapt to the words it doesn’t yet know. The implication for us here is that we have to develop solutions that are very dissimilar from one another. It is because of this that we have such high hopes for systems built on AI technologies, which in theory would be easier to upscale and more adaptable to our requirements.
Projects based on speech synthesis are already yielding promising results. What with all the modern developments, it’s not a problem for us to synthesize speech to a high standard. The main challenge here is enhancing it with emotional coloring.
Finally, another crucial aspect to improve upon here is the elimination of structural glitches. The users we work with could find themselves in a wide range of locations: inside a building, outside in the noisy street, wherever. We have to address this to make our technologies process only the necessary information.
As for chatbot systems, there is a topical issue of effective identification of facts and objectives the user wants to achieve when referring to the company. What’s crucial here is for the chatbot to grasp the context of the inquiry, and it is one of the most complex problems we face when developing such systems.
Projects conducted with ITMO University
Our company has a special department aimed at interacting with universities. It is coordinated by Maksim Gashkov, and it was him who was responsible for establishing contacts with ITMO University specialists, including Alexander Boukhanovsky (Director of the School of Translational Information Technologies – Ed.) We made several official visits to the University, during which we identified several promising areas for future cooperation. This work resulted in two joint projects.
The development of a system of speech recognition, generation and analysis of noise (led by Dmitry Muromtsev, Head of ITMO University’s international laboratory ‘Information Science and Semantic Technologies’).
Dmitry Muromtsev’s team has a wealth of experience in the field of NLP and all sorts of ontologies. Our joint project is practice-oriented and aimed at creating a software-hardware complex that would allow for mass testing of speech recognition and analysis systems and be applicable for working with noise.
The development of a personalized medical recommendation system (led by Sergey Kovalchuk, a senior researcher at ITMO University’s eScience Research Institute)
Sergey Kovalchuk’s team conducts research in the field of medicine. Our joint project is directed at the analysis of open textual data garnered from various sources. The main defining feature of this project is that the specialists work with unstructured data, which isn’t the simplest task to do. Our primary objective is to develop ways for processing this data to accurately retrieve different topic-specific facts and figures.
The results of this work will be then integrated into our decision making support system; we also plan on exploring further personalization of the solutions we offer. There are of course standardized treatment trajectories and protocols out there that all patients should follow, but each and every one of us is unique, and that explains why the most up-and-coming trend in contemporary medicine is the personalization of medical care. This requires searching for the facts that diverge from the common pattern, which, in turn, leads to a more effective treatment.
But I want to emphasize that our role is not to offer this treatment but, rather, to assist a patient in understanding where it is that they have to seek help. In other words, our main goal is routing on the local level. I hope that working with ITMO University on this joint project will allow us to solve a large number of tasks we need to tackle to achieve this goal.
These projects are as important for us as they are for the University, as they provide students with new research topics, and give them the opportunity to contribute to the solution of complex scientific and technical tasks, as well as to gain valuable experience of working with a major industrial partner.
Specialists needed in hi-tech companies
As of now, our group consists of 40 specialists, but there is a plan for a 2.5 increase in the number of employees next year.
Firstly, we at the functional AI group of MTS Innovations Center would be happy to welcome students from the first year of studies and older. We offer internship positions with the workload starting from 30 hours per week.
We’re also on the look-out for both developers and project leads skilled in the processing of texts based on different machine learning approaches. Also crucial are specialists with expertise in the field of sound processing, for instance, speech recognition and synthesis, as well as in the field of recommendation systems.
The most sought-after on today’s labor market are specialists capable of switching between different tasks in a specific field, and, most importantly, of formulating these tasks accurately and correctly. There are but a few universities that train such specialists. I’m convinced that ITMO University is one of them; this is clear from the two projects we’re carrying out with ITMO.
Mikhail Mishurovsky has been working in the field of new technologies for over ten years. From 2007 to 2018, he worked in Intel and Samsung Research Center in Moscow, where he conducted development of algorithms for next-generation video coders, including those based on the ML/AI technologies, and contributed to the development of AI-based solutions for analyzing video stream quality and improving the effectiveness of video coding. As of now, Mikhail works in the functional AI group of MTS Innovations Center, where he oversees the development of technologies for the processing of data, including natural language data, obtained from different sources, on the basis of machine learning methods.