29.07.2024‘The conference featured a new track, position track designed for papers that focus on viewpoints or trends within the ML field rather than presenting new research findings. ICML 2024 hosted a number of social events that provided networking opportunities. This year, 9,473 submissions (excluding desk-rejected papers) were received by ICML, representing an increase of 44% from last year. Out of these, 2,609 submissions were accepted for presentation at the conference, resulting in an acceptance rate of 27.5%.’, says Kamil Adamczewski.
Kamil Adamczewski / Michał Krutul, Sebastian Jaszczur, Jakub Krajewski, Jan Ludziejewski, Kamil Adamczewski (all IDEAS NCBR)
‘Given the works presented at ICML, we can distinguish a few trends. LLMs are the most popular field of research. While building larger models produces better results (as presented in our paper “Scaling Laws for Fine-Grained Mixture of Experts”), more and more focus is on building small, good-performing, efficient models,’ adds Kamil Adamczewski.
‘It was my second time at a premier ML conference, and again, I was not disappointed. Although most of presented research was on LLMs and reinforcement learning, which is slightly out of my area of expertise, I still found a lot of new interesting papers from my own area of research and was happy to connect with their authors,’ says Marcin Przewięźlikowski.
‘We had an opportunity to listen to talks by experts such as Chelsea Finn, who presented exciting directions for efficient and embodied Machine Learning – something which is very close to my research interests. Another interesting point was a talk by Lucilla Sioli – the European Comission’s Director for AI and Digital Industry, who presented the vision for AI development in Europe,’ adds Marcin.
Bartłomiej Cupiał (IDEAS NCBR) / Aleksandra Nowak (formerly IDEAS NCBR), Łukasz Gniecki, Filip Szatkowski (IDEAS NCBR)
