A group of young scientists in Armenia have adopted a new hashtag: #ՀզորԳիտությունՀզորՀայաստան (#PowerfulSciencePowerfulArmenia). They are using it to publish live videos on their personal Facebook pages of “virtual tours” through their laboratories and interviews sharing their viewpoints about the state of science in Armenia. The social media campaign aims to bring attention to the science sector with its problems, needs and potential.
As mentioned in a recent public forum held at the National Assembly, state expenditures on scientific activities have not exceeded 1.2% of the state budget for decades. However, the Law on Scientific and Scientific-technical Activities suggests that state expenditures on science [technical expenses, grants/competitions, publications, research, etc.] should be increased in parallel with the country’s economic growth. While Armenia has mostly recorded steady economic growth throughout the past decades (up to 14% GDP growth), public expenditures on scientific activities have not increased adequately. State funds dedicated to this sector have typically been about 0.25% of Armenia’s GDP, while a typical rate in most other countries is around 1%. Squeezing financial resources has resulted in poor compensation for scientists. As a result, much of the talent pool either left the country to continue their academic career or moved to the private sector, where salaries are much higher. As a result, academia is no longer appealing to the young generation, when they are choosing their career path.
Regardless of the insufficient resources, there are still several devoted scientists, institutions and labs that continue growing and promoting academic life thanks to international grants or donations from the private sector. YerevaNN – an Armenia-based non-profit computer science and mathematics research lab – is among those labs. YerevaNN’s story stands out because its focus area – machine learning – is relatively new to Armenia and the world. The team started everything from scratch. “Judging from the number of scholarly articles published in the field of computer science by Armenia-based scientists, we can assume that this field of science is not well-developed in Armenia. It’s much weaker than, for example, physics or bioscience, which are in a comparatively better state in Armenia,” says Hrant Khachatrian, the director of YerevaNN and one of the young scientist activists.
Khachatrian shares the story of how YerevaNN got to stand on its own feet.
Everything started in a classroom at the department of Applied Mathematics at Yerevan State University. Khachatrian and a few of his peers would get together weekly to watch tutorial videos about neural networks. “They were only hour-long videos, but we would take two hours to watch them because we would pause them regularly to discuss the subjects with each other and make sure we were all on the same page,” explains Khachatrian. With a shrinking circle of interested students, Khachatrian and a few other friends eventually decided to create an environment to come together and hold discussions more regularly. Moreover, they decided to turn it into an academic environment where students and researchers would not only learn and experiment, but also uncover new potential and work on scientific innovations.
“A few of my senior friends highly encouraged and supported us in this endeavor. Hence, we established a foundation, and those very friends formed the Board of Trustees,” remembers Khachatrian. The senior friends he is referring to are leading figures in the machine learning (ML) scene in Armenia: Gor Vardanyan, a co-founder at Chessify, one of the growing Armenia-based ML startups, Vazgen Hakobjanyan, a partner at SmartGate VC, which invests in ML startups, and Ruben Meschyan, a founding software engineer at Cambridge Semantics.
“We didn’t elaborate much on what to call our foundation, we just needed a title to put our code on GitHub. Since most of the neural network innovations are named with an abbreviation ending with ‘NN’ for ‘neural network,’ we decided to simply call ourselves ‘YerevaNN’ as an ML research lab based in Yerevan,” explained Khachatrian. Next, Khachatrian and the team rented a small space close to Yerevan State University, which would be comfortable for students to attend.
With this, YerevaNN was officially launched in 2016. Khachatrian and the team quickly realized that their path would be full of challenges. The first one: they couldn’t find a scientific supervisor. Every scientific research group needs supervision by professors, who keep up with scientific developments in their field. “Unfortunately, we don’t have them in Armenia after the 1990s in the field of machine learning specifically,” shared Khachatrian.
It is no secret that science in Armenia has been in a coma for the last 30 years. The younger generation did not step in to replace the old one in order to continue academic work. Most postgraduate students, who were supposed to take on the role of supervising a new generation, either left the country or went into industry, where salaries are much higher. “Sometimes, I think that the absence of industry [in Armenia] in physics or bioscience saved those fields because the talent stayed in academia and the transition from their generation to a younger one went smoother,” elaborates Khachatrian.
Leveraging the Diaspora
With no local supervisor, the YerevaNN team started to reach out to Armenian scientists working and living abroad. They found many successful professors and scientists following academic developments in computer science and its different fields, who were ready to provide support. In one of his recent articles, Khachatrian highlighted the scholarly potential of the Armenian diaspora, mentioning several scientists in different scientific fields, whose articles were published in high-ranking international scientific journals and conferences.
One of those scientists is Dr. Aram Galstyan, the Research Director for Data Science and Machine Learning at the Information Sciences Institute of the University of Southern California. He was one of the early supporters of YerevaNN, and has helped the team by supervising research projects, connecting them to other scientists or post-doctoral students and in many other ways. As Khachatrian explained, these scientists usually have many ideas and scientific problems for research, but because of a shortage of time, they have to pick only a few focus areas to work on. Some of them, however, decide to assign those problems to other research teams with promising potential and supervise them throughout the research process. As a result, they not only get to work and innovate in an area they wouldn’t otherwise find time for, but they also develop and support younger researchers, transferring their knowledge and experience to a new generation. YerevaNN became one of those research teams for the University of Southern California (USC) and it continues to cooperate with them by providing scientific research as a service. This helped them to secure alternative financial resources, other than industry donations, and find a path to financial stability.
Starting with this partnership, YerevaNN continued its cooperation and joined up with other universities and new supervisors. Their members have attended, presented and lectured at world-renowned forums, conferences, workshops and other events in dozens of countries. Moreover, Khachatrian is now involved in organizing a major event dedicated to machine learning, DataFest Yerevan, in April 2020, where they will unite their networks in Armenia for new connections and knowledge sharing.
Making a Splash
After three years of hard work, YerevaNN achieved one of its most-awaited accomplishments: they eventually got their first scholarly publication in an important scientific journal and high-profile conference in 2019.
The first one was a scholarly article entitled “Time Series Prediction on Medical Data,” which was one of the first topics studied at YerevaNN, suggested by David Kale, a PhD student at USC, who supervised this research. In this research, they used machine learning to analyze medical data for not only treating diseases but also making a prognosis on a patient’s data, up to mortality prediction. It was published in Nature Scientific Data, a leading journal in the sector. A year earlier, they posted the “non-published” version of the paper on arxiv.org, a free-distribution and open-access platform for scholarly articles, and gained massive attention by researchers and scientists from all around the world. Recently, this paper passed the 100 citations mark on Google Scholar. Secondly, YerevaNN successfully presented a paper at the BioNLP workshop (ACL’19). Here, they presented their research of machine learning applications in bioscience entitled “Relation Extraction From Biological Literature,” supervised by Aram Galstyan. In this study, they created a well-balanced database of biological interactions from texts to enable automatic extraction of biological relations from scientific publications.
Such scientific publications are very important accomplishments in the scientific process since they validate scientific innovation. “Here, it’s important to understand the scientific process. Scientists submit their articles for review to related science conferences or journals. In case of a positive review, they get published and in case of a negative review, they get detailed feedback, which will help to improve their work so they can resubmit later,” explained Khachatrian adding that reproducibility has a very important role in science when innovations are repeated, tested and reviewed. Any completed work should be reproduced by other scientists to prove that everything is right and there’s no forgery. Only after that can the suggested innovation be considered valid as it’s been proven by different independent groups of scientists.
Such publications involving Armenia-based scientists bring more attention to the scientific potential of Armenia. Khachatrian suggests this model of YerevaNN be applied to universities as well: find supervisors from leading universities and form research groups engaging students at all levels to work on the suggested topic. As Khachatrian mentioned in his article, there are many supportive scientists in the diaspora, who are ready to help in growing the science potential in Armenia, and the case of YerevaNN research lab is proof. This model might not be applicable to all branches of science, especially to experimental sciences, which require expensive equipment and materials, but the more theoretical disciplines can benefit from such collaborations.
Industry-Science Collaboration Model
They say that scientific work needs to become a business to bring value. However, this is only a small part of the value science can bring to the industry sector. “We do not need science for their product algorithms or innovative solutions, we need them in the industry for their scholarly mind and knowledge,” Kachatrian explains. “The industry needs solutions to their problems and science needs yet unsolved problems regardless of the product and the business application.” Hence, the major effect that research groups can leave on the industry is developing high-quality personnel. This means that industry won’t need to headhunt scientists, leaving their spots in academia vacant, but outsource their problems to those research groups and get solutions with lower costs and better results. With this model of “outsourcing scientific problems,” a research group can continue to train students into new scholars. Also, research groups can use their proceeds from industry to achieve financial sustainability. That being said, students won’t have to leave academia in search of higher salaries and a better work environment. Of course, not all students will stay in science, but the reproductive system will help to fill the personnel gaps both in academia and in science.
To sum up, although financial resources are crucial for science development in Armenia, there are other important gaps that money won’t fill on a short-term basis. Scientists’ plans are more long-term as they cannot predict how long their research will go, so it’s not easy to persuade them to change their work or country. Hence, the solutions related to science should also consider long-term approaches. With the academia-industry and “research as a service” cooperation models, scientific research groups can solve these most important problems by nurturing new generations of scientists, keeping the talent in academia and solving scientific problems in the industry. Fortunately, Armenia has a big advantage here – the Armenian diaspora can both connect and supervise a new generation of scientists in Armenia. YerevaNN’s story and plans show that it can become possible, at least for some branches of science, when there is a clear focus and strategic model.