Radical Reads: A conversation with a Ukrainian AI researcher

Editorial Team


Dr. Dmytro Mishkin

Image Source: Проект «Атоми», The Ukrainians

Ukraine’s research and startup community plays an important role in the global, interconnected AI ecosystem. Following the Russian invasion of Ukraine, Radical Partner Aaron Brindle reached out to Dr. Dmytro Mishkin, a Ukrainian AI researcher and ex-startup founder. Dmytro is a member of the Expert Committee on Artificial Intelligence at the Ministry of Digital Transformation of Ukraine. He is also the co-founder of the Eastern European Computer Vision Conference and started the Ukrainian Research Group “Szkocka,” an initiative to promote cooperation between Ukrainian researchers and supervisors doing high-quality academic research. He is the core maintainer of Kornia – a popular computer vision library. Aaron reached Dmytro in Prague, Czech Republic. The following is an edited version of their exchange.

Aaron Brindle (AB): How are you doing – are you and your family safe?

Dr. Dmytro Mishkin (DM): My immediate family – my wife, my daughter – are safe. We are in Prague right now. My parents, however, are not safe. They live 200 metres from where there was heavy shelling.

AB: What are you hearing from your colleagues in the Ukraine AI community this week?

DM: Many of my friends are helping the Ukrainian army to defend our country. My colleagues are also very keen to spread the word about Russian aggression. But, they are also trying to do their usual work. They are still doing their research and, as much as it is possible, developing and delivering products for their companies. 

AB: Billion-dollar Ukrainian startups, including Grammarly and GitLab, are now global brands and they still maintain AI offices in Ukraine alongside the AI R&D labs of global tech giants such as Samsung, Google, and Rakuten. According to the 2020 Oxford Government AI Readiness Index, Ukraine was the number one artificial intelligence provider in Eastern Europe. What can you tell us about the AI community in Ukraine?

DM: The Ukrainian AI community is indeed very strong and mostly industry-based. Two major startups in Ukraine were recently acquired, including Augmented Pixels by Qualcomm and Apostera by Harman (Samsung). AI research is also being undertaken in Ukraine’s enormous agricultural industry. One of my friends founded Kray Technologies which develops drone-based crop protection solutions. They do a lot of R&D in computer vision and machine learning. 

There is a powerful emphasis on research within Ukraine’s startup community. Even our IT outsourcing companies like ELEKS and SoftServe publish at CVPR. Beyond startups, Ukraine has a thriving AI academic community, led by Ukrainian Catholic University, Kyiv Polytechnic Institute and many more. 

AB: Can you tell me a little bit about your computer vision research and the kinds of applications you’re working on?

DM: My research is about finding correspondences between images of the same scene. It is used in 3D reconstruction, visual localization (think of the Google Maps mode, where you open the camera and get the direction where to go, based on what the camera sees), augmented reality and many others. Basically, any application where you need to know your camera position and orientation. You can check out my talk at the EECVC conference for more details. Also, many of the algorithms I developed are available in Kornia, a popular computer vision library, which is downloaded around 300,000 times per month. I am the core maintainer of it.

AB: Can you tell me about your proudest moment as an AI researcher?

DM: My proudest moments were probably when my algorithms were added to the Stanford course on convolutional neural networks and to Richard Szeliski’s classic textbook on computer vision.

AB: Thank you for speaking with me, Dmytro.

DM: Thank you.



5 Noteworthy AI and Deep Tech Articles: week of February 28, 2022

1) Who Is Behind QAnon? Linguistic detectives find fingerprints (The New York Times – subscription may be required)
Computer scientists have identified the likely authors of the “Q” account behind the QAnon movement. The researchers used machine learning to compare subtle patterns in texts that a casual reader could not detect. The software broke down the Q-texts into patterns of three character sequences and tracked the recurrence of each possible combination. Machine learning is advancing stylometry and has cracked famous cases in linguistic forensics. For example, AI helped reveal that J.K. Rowling wrote the 2013 mystery “Cuckoo’s Calling” under another pen name. In recent years, AI techniques have helped detectives in the US and UK solve murder cases involving forged notes and text messages. 

2) Winter wheat yield prediction using convolutional neural networks from environmental and phenological data (Nature)
Climate change and an increasing need for food security is driving demand for accurate crop yield prediction models. Scientists from the US and Germany found nonlinear models to be more effective in finding the functional relationship between the crop yield and input data. Crop yield forecasting depends on many interactive factors, including crop genotype, weather, soil, and management practices. The study analyzes the performance of machine learning and deep learning methods for winter wheat yield prediction using an extensive dataset of weather, soil, and crop phenology variables in 271 counties across Germany over the last twenty years. 

3) Good news about the carbon footprint of machine learning training  (TechRxiv)
Four key practices can reduce the carbon footprint of machine learning workloads by large margins. The proposed “4Ms” – model, machine, mechanization, and map optimization are outlined in a preprint paper by Google researchers accepted for publication in IEEE Computer. The researchers argue that the 4Ms, when used together, can reduce energy by 100x and operational carbon emissions by 1000x. The paper also recommends that machine learning papers include emissions explicitly to foster competition. The researchers suggest that, if widely recognized, the 4Ms could create a virtuous circle that will bend the curve so that the global carbon footprint of machine learning training begins to shrink. 

4) This AI could be robocallers’ kryptonite (IEEE Spectrum)
Robocallers placed nearly 4 billion nuisance calls in the US in January 2021. US Government efforts to combat the calls have been futile. Research teams are now employing machine learning to cull nuisance calls. An app developed at Georgia Tech picks up your incoming calls and screens them with deep learning modules to determine if the call is unwanted. The virtual assistant achieved a 97.8 percent success rate at pinpointing robocalls, and its creators hope to deploy it as an app. Another machine-learning-based system, created for Chinese mobile phones by researchers at Shanghai Jiao Tong University and the University of California, Berkeley, achieved a similar 90 percent success rate.

5) AI machines have beaten Moore’s Law over the last decade, say computer scientists (Discover)
Since the 1990s, computer scientists have measured the performance of the world’s most powerful supercomputers showing supercomputing performance increasing in line with Moore’s Law. Deep learning techniques led to a step-change in computational performance, but their impact has not been as well ranked. Computer scientists at the University of Aberdeen have developed a method to measure AI systems’ performance. From 1959 to 2010, the computational power used to train AI systems doubled every 17 to 29 months. But in the last ten years, what the researchers call the “Deep Learning Era,” progress has been rapid. The overall trend speeds up and doubles every 4 to 9 months, outperforming Moore’s Law.  

Share this edition of Radical Reads: 

Share on facebook
Share on twitter
Share on linkedin

Radical Thinking: AI’s Next Wave

At Radical we have seen hundreds of startups looking to shape the future of AI technologies. From this experience, I developed a framework for categorizing different waves of AI adoption. Understanding the changes that have come before situates the current moment and helps companies plan for the future.

Read More »

© 2022 Radical Ventures Investments Inc.