Scientific discoveries and results don’t achieve their true value unless we empower others to use them effectively
It was a seemingly ordinary summer’s morning in 2006. The then twelve-year-old me woke up early and sat down at the computer, wanting to complete a level in some game—as I had done on many a previous morning. My mum burst into the room and declared loudly that I had absolutely no chance of getting into the Mathematical Grammar School with such an approach! And yet, this day was not quite ordinary: I had to take the entrance exam for the MGS’s 7th grade experimental class in just a few hours. Perhaps I had more luck than smarts, as I had major issues with the material that was taught at MGS over the following years—especially geometry—but my actions with that exam changed my life completely.
That wasn’t the only day that marked a turning point in my career, but it is the one I remember most vividly. Subsequent moments – my entrance exam for Computer Science undergraduate studies at Trinity College, University of Cambridge (where I received a full scholarship), the day I chose to leapfrog a Master’s degree and embark on a Ph.D. in Artificial Intelligence at Cambridge, without any preparation, or the days I decided to send unsolicited emails to some of the most influential scientists seeking collaboration – further shaped me into the scientist and person that my colleagues and friends know me as today.
Teaching is a simple way to gain superpowers: explaining complex topics to others is the best way to explain them to yourself
My job entails working on artificial intelligence on graphs. Most modern AI systems work on simple data, such as images, text or sound. However, data from nature aren’t simple and often have an irregular structure. My path in researching them has led me to exciting achievements: a system that predicts travel times in Google Maps, helping mathematicians discover hidden structures in complex objects (a paper that graced the cover of Nature, the world’s most prestigious scientific journal) and, most recently, collaborating with Liverpool Football Club on the development of the first AI system capable of providing useful suggestions to football coaches.
My development path hasn’t been flawless; I’ve made mistakes at almost every step. When applying to Cambridge, in conversations with senior colleagues from MGS who’d successfully passed this stage, I learned firsthand how to make a strong application without repeating their mistakes. This inspired me to help new generations improve and to pass on my experiences whenever possible. My first step towards this goal was organising the Week of Informatics at MGS, an initiative through which MGS alumni share their experiences and knowledge of computer science with current pupils. I later became an Affiliated Lecturer at Cambridge University, where every year I convey the wonders of geometric deep learning to Master’s students.
If you recall how I began this text, geometry never came easily to me. I only developed an intuition for this field when I put myself in the position to have to teach it to students. Teaching is a simple way to gain superpowers: explaining complex topics to others is the best way to explain them to yourself. Perhaps that’s why I insisted so much on bringing this year’s prestigious Eastern European Machine Learning Summer School (EEML) to Serbia, where, over the course of week, we did our best to present the latest AI trends to the local academic community in Serbia and give them the “wind in their sails” to realise that working at such a level isn’t beyond them!
