Revisiting the highlights of the year in a Retrospective article has become a small tradition of our group, and as people that honour traditions, we revisit the year 2023 from the lenses of our group. Let's go!
The year 2023 finds the DMML group at its record high in terms of numbers of team members and ongoing projects.
The team is currently composed of 14 members.
During the year 2023, the team was strengthened with the addition of four members, while we greeted Naoya Takeishi who left for Japan, with a faculty position waiting for him at the University of Tokyo. The four new members are the following:
Dr. Lionel Blondé returned in action in our team, providing his expertise in both applied and fundamental part of our team's research. With long standing expertise in RL, Lionel has been a great addition to our team, in terms of expertise and spirits!
Singh Gurjeet, or simply Jeet for friends, joined our group this year as a new PhD student. Before joining, Jeet finished his M.Sc in Data Science at the University of Padua (Italy). Jeet is an excellent addition to our team's dynamics, gracefully skilled in both hard and soft skills. His research focuses on Simulation-Based Inference (SBI) and spans Generative Models and Optimisation problems, while in his free time he likes to run, hike, and play futsal! A football lover was long awaited in our group! You may learn more about Jeet, by visiting his personal web page.
Khoa Nguyen
, joined our group soon after Jeet, as a PhD student. He holds a B.Eng and a M.Eng in computer science and information systems at IMT Atlantique (France), while he has worked as a Jr. Data Scientist at Trusting Social. Khoa has a wide range of research interests. He is currently looking into the development of new generative models on graphs and studying their particular applications to accelerating drug discovery. He is a calm force and the team enjoys his aura!
The most recent addition to the team is Giangiacomo Mercatali (aka Gian), who will soon share more info about his background (more info in retrospective 2024!).
As we have numerous new research projects (that we present right below), we will soon be advertising new positions at PhD and post-doc level!
The completion of 2023 finds our team with the most variable palette of research projects so far, with 8 research projects!
There were 4 ongoing projects that started earlier on and that continue throughout 2023:
- EO4EU: a Horizon Europe project, related to Earth Observation, which aims to provide innovative tools that will enable a wide spectrum of users to benefit from EO data. Our team contributes with providing label-efficiency through self-supervised learning methods, and by designing tailored methods of learned compression. More info, on the project's website eo4eu.eu
- Automated Bridge Defect Recognition: an Innosuisse project, focusing on bridge inspection and ML-driven damage detection.
- Learning generative models for molecules: An SNSF project on generative modelling within discrete structures.
- MIGRATE: An SNSF Synergia project that deals with three complementary domains: geology, seismology and machine learning.
In 2023, the team got 4 new research projects!
- CoORDinates: Funded by the Swiss Open Research Data Grants (CHORD) of SwissUniversities, CoORDinates explores the existing landscape of Open Research Data (ORD) within the Indoor Positioning community, aiming to propose relevant guidelines, leveraging experiences of existing ORD practices of other fields.
- Interpretable Condition Monitoring for Complex Engineering Systems: In this project we explore the idea of grey-box (hybrid) modelling, where data-driven machine learning models and theory-driven expert models are combined. The project is funded by the Swiss National Science Foundation under the Strategic Japanese-Swiss Science and Technology Program. It is an international collaborative project with a counterpart in Japan; we are working with the Artificial Intelligence Lab of the Research Center for Advanced Science and Technology, the University of Tokyo.
- Metathesis: With a full title "Modelling pathological gait with machine learning for treatment selection support", Metathesis continues a long standing collaboration with the Kinesiology lab of Prof. Stéphane Armand (University of Geneva/ Geneva University Hospitals). The project is funded by the special call on Health and Wellbeing, directed towards researchers of UAS and UTE by the Swiss National Science Foundation.
- HYPER-AI: A Horizon Europe project (currently in Grant preparation), funded under the HORIZON-CL4-2023-DATA-01-04 topic on the Cognitive Computing Continuum. In this project, our group will explore Multi-Agent Deep Reinforcement Learning Methods for Data Management.
Apart from exploring new lands, our group aims at deepening our gained expertise in certain research lines.
- Molecules/graphs: The SNSF project Learning generative models for molecules extends the ways previously explored in the SMELL project, working on molecular structures from a different perspective.
- Grey box modelling: The project Interpretable Condition Monitoring for Complex Engineering Systems continues the research line on grey-box (hybrid) modelling, explored in different contexts, as in SimGait and in relevant publications.
- Human gait analysis: Metathesis is the most mature stage of our human gait modelling research line. After the introductory project and the reality check of the Synergia project Simgait, Metathesis opens concrete and well thought research directions for the next four years.
- Self-Supervised Learning (SSL) for sample-efficiency: In the EO4EU Horizon project and in the Automated Bridge Defect Recognition Innosuisse project, we explore SSL and its capabilities from different viewpoints.
- Reinforcement and Imitation Learning: In Metathesis we will capitalise on the RL and IL expertise obtained in Simgait, while in HYPER-AI, a multi-agent RL approach will be put in action!
- Indoor and Outdoor Positioning: The particular interest of Greg for this research line, brought his third relevant Grant on these topics. After LiBertaS which was a natural continuation of Eratosthenes, dealing with sample efficiency for positioning datasets, CoORDinates takes a turn towards methodological aspects: FAIR-compliant Open Research Data, Reproducibility, Research Transparency, etc.
Other projects though are invitations to new journeys, where the team is invited to exploit ML machinery in novel grounds!
In terms of publications, our team members published the following works this year:
- Frantzeska published her work "Back translation variational autoencoders for OOD generation" in the ICLR 2023 Workshop on Domain Generalization.
- Naoya continued his research on Grey box Modelling, with the work "Deep Grey-Box Modeling With Adaptive Data-Driven Models Toward Trustworthy Estimation of Theory-Driven Models" that was published in the International Conference on Artificial Intelligence and Statistics.
- Yoann produced two novel works this year:
- "Graph annotation generative adversarial networks" which was published in ACML, and
- "Vector-Quantized Graph Auto-Encoder" a preprint that continues Yann's work on Graph Neural Networks (GNNs).
- Joao's paper "Sample-Efficient On-Policy Imitation Learning from Observations" further evolves his work on Imitation Learning and Sample Efficiency.
- Maciej's "Calibrating Neural Simulation-Based Inference with Differentiable Coverage Probability" is a great contribution to this year's NeurIPS!
While in all these works we mentioned the leading authors, we should acknowledge of course the contributions of all co-authors. Apart from Alexandros who is the common denominator of our team's efforts in all these works, as co-authors we congratulate: Naoya, Imhan, Lionel, our alumni Magda Gregorova who is now heading CAIRO, and our colleagues Antoine Wehenkel, Arnaud Delaunoy and Gilles Louppe from the University of Liège!
Last but not least, Imhan also saw his work with his teammates of his former group reaching the publication stage:
- "Advancing Generative Modelling of Calorimeter Showers on Three Frontiers" is another contribution to NeurIPS 2023, at the specialised workshop 'Machine Learning and the Physical Sciences'
- "L2LFlows: generating high-fidelity 3D calorimeter images" is a contribution published in the Journal of Instrumentation
As far as mobility is concerned, our members of the DMML group have been on the move over the last year, participating in different activities such as: summer- and autumn- schools, international collaborations and invited talks.
In terms of Institutional activities, our team members contribute to our school's life from various viewpoints. Apart from the numerous courses and Theses on Bachelor's and Master's levels, most of which (unsurprisingly) revolve around Machine Learning, our team is very active in contributing to the participatory bodies of our University.
Alexandros has been elected to serve in the 'Conseil représentatif HES-SO Genève', and Greg has been elected to serve at the 'Conseil Académique Haute école de gestion Genève' at the local level of our University, as well as at the 'Conseil de concertation HES-SO' and the 'Conseil participatif du domaine Economie et Services HES-SO' at the inter-cantonal level of our University.
Overall, our team is serving our University by holding four seats in its participatory bodies, demonstrating an investment in the advancement of our Institution!
This was in a few (or a bit more) words the year 2023 for our group.
We are happy to continue this tradition for the fifth consecutive year. You may revisit the retrospectives of 2019, 2020, 2021, and 2022 to gain a view of the team's evolvement through time!
We are looking forward to a fruitful and enjoyable new year. May it bring the most pleasant vibes of collaboration and knowledge discovery!
Best wishes for a loving and creative new year!
