Bitdefender at EEML2019
This summer Bitdefender’s Machine Learning Research Team participated in the second edition of the Eastern European Machine Learning Summer School, this year’s highlight event for our fast growing machine learning community in Romania.
What is EEML?
The EEML Summer School gathers top researchers from universities and companies, offering an intensive week full of lectures and practicals to participants ranging from undergrad students to post-doc researchers.
The summer school was successful in its mission to connect Eastern Europe with the rest of the world, welcoming attendees from 42 different countries. Being true to the schools values on inclusiveness, women represented almost a third of the participants, more than the usual ratio encountered in STEM.
Furthermore, at the national level, EEML offered the chance to meet and strengthen the collaboration between research hubs in Iași, Cluj-Napoca, Timișoara, and Brașov.
I like the fact that, although the school grows visibly in size, you still have a chance to talk and share opinions with speakers and other participants.
Bringing everyone together for a full week makes EEML an ideal place to exchange ideas with others working on similar topics, to draw inspiration from others’ approach to their problems, and this adds a valuable extra to the overall learning experience.
It's been more than two months since I participated at EEML but the ideas floated there and the relationships I started still seem highly relevant to me - as should be the case for any good summer school. I fondly remember the things I learned from David Nagy (@dvgnagy) on Bayesian reasoning, the discussions I had with Rob (@RobertTLange) on RL, multi-agent systems and much more or the reading suggestions on causal inference I got from Jeroen Berrevoets (@J_Berrevoets). EEML also facilitated starting a research collaboration with Błażej Osiński and Maciej Wołczyk after discussing it over some beers.
Bitdefender among organizers
EEML’s mission was laid out by a group of successful romanian researchers that wanted to ensure eastern european talent pool gets access to the same resources as the more established academic centers in the western world. Doina Precup, Răzvan Pașcanu, and Viorica Pătrăucean were the leading organizers of the summer school, and their dedication made everything possible.
Bitdefender was one of the Gold Sponsors of EEML proving its commitment to advance the local research community and invest in local education. Beside that, our own Elena Burceanu was one of the organizers of EEML, putting a lot of time and effort into making sure that every little detail will happen like clockwork.
EEML is a huge event so it was all hands on deck. Therefore we put the volunteer yellow shirts on, took a deep breath, a sip of coffee and joined the efforts. Our colleagues helped participants finding their way from the airport to the accommodation, prepared the welcome bags, they made sure every video projector works, and that there are spare batteries for the microphones in the lecture room.
The summer school kicked off with a humorous presentation about the Romanian culture from our own Florin Brad. That moment provided everyone with the energy to go through the full week.
In Florin's own words:
This year I held a five minute presentation about Romania during the opening of the summer school. They say the best way to learn something is to explain it to other people, so I hope people have learned one or two things about our country (because I sure have).
The lectures
The lectures at EEML covered a broad range of topics in Machine Learning aiming to both build the theoretical foundations for the newcomers and also hint to the research frontiers in the domain.
Furthermore, the diverse background of the speakers enabled a wide variety of lectures, ranging from pragmatic descriptions of the latest research results to theoretical crash courses in the different areas covered by the summer school.
Posters
EEML participants were encouraged to present their current work in a poster session that spanned over two evenings (actually well into the night). That was a great opportunity for them to get feedback from the seniors.
It was a great opportunity to present my work at EEML, getting a chance to share opinions with amazing people with varied expertise in computer vision. This experience had a positive impact and helped me shape my next research steps as many discussions brought a new perspective over my work.
Our team unveiled three internal research efforts. Iulia Duță and Andrei Nicolicioiu presented their novel recurrent space-time graph neural networks for video understanding (also accepted at NeurIPS!!!). Emanuela Haller discussed her work on spacetime graph optimization for video object segmentation. Florin Gogianu introduced his idea to use uncertainty to prioritize past experiences for training agents through reinforcement learning.
EEML also provided me with the opportunity of presenting my work on Bayesian-inspired prioritization measures for the Experience Replay buffer during one of the two poster sessions. It was actually my first conference-like poster session I attended so it was a rather interesting experience. I received some great suggestions from Diana Borsa which might help me polish this research further and also got some follow-up questions from the people I met there. Somehow I also got one of the Best Poster Awards and that wrapped-up nicely the great experience I had attending EEML.
Teaching assistants
EEML practicals were challenging applications of the theory discussed during the lectures. The subjects were quite diverse covering Image classification, reinforcement learning, recurrent neural nets for language models, or generative models such as variational autoencoders. The teaching assistants had the job to guide the participants into solving this tasks, and leave the summer school with a consistent experience in training deep neural models using Tensorflow. The TAs were researchers from prestigious universities, and companies such as DeepMind, and Bitdefender.
For me, it was quite a different perspective this time, as I join the summer school as a teaching assistant, an attribution that I loved to hold as it gave me the opportunity, besides captivating lectures held by top researchers all over the world, to interact more with participants and other TAs. It was a great mix of receptive students, awesome devoted colleagues and a strong team behind all the labs that prepares everything to be fun and valuable._
Our own colleagues, some of which are also teaching at University Politehnica of Bucharest or University of Bucharest (Iulia, Florin, Andrei, Ștefan, Ema, Tudor) got involved in guiding the participants through the new stuff in each laboratory.
I am a teaching assistant at the university, but here it was way more challenging given the hard topics and the smart and enthusiastic people attending. Brilliant researchers were also acting as teaching assistants and I really appreciate interacting with them during and before the labs. It was very interesting to see how other TAs explain the topics they are experts on.
Epilogue
EEML was a huge opportunity for Bucharest to join and interact with the relevant actors in the machine learning community. Bitdefender is proud to be so deeply involved in bringing this event to our city.
Facilitating new relations between like-minded researchers at the beginning of their careers is not the only thing making EEML great. I felt I gained a lot by participating or assisting in discussions with researchers like Răzvan Pașcanu, Doina Precup, Andrei Rusu or Akata Zeynep. Simply hanging around these people and hear them discussing any topic for 20 minutes can leave you with tons of insights and problem formulations they've deeply thought about and to which you could not have access to otherwise.