As we approach the Christmas period, and holidays are ahead of us, we seize the opportunity and take a moment to reflect on the year that is about to be completed. This yearly retrospective is becoming a thing in the dmml group, as we had one in 2019 and in 2020 as well.
So, let's go!
This year Maciej Falkiewicz joined our group. Maciej is an enthusiastic PhD student that adapted in the team in zero time, and it feels like he is with us since years. Maciej's spectrum of research interests is wide, spanning from invertible neural networks to physics integrated machine learning and beyong! He is always enthusiastic about informing the team about the latest published innovations of ML. We are delighted to have him in the group!
The year 2021 has been a year of several Theses defences! More particularly:
- Jason Ramapuram, who had started working in Apple research last year, defended his PhD Thesis entitled "Finding signals in the void: Improving deep latent variable generative models via supervisory signals present within data."
- Amina Mollaysa defended her PhD Thesis entitled "Structural and Functional Regularization of Deep Learning Models", and is working as a postdoctoral researcher in our group. She continues her research while she also teaches the course on Unsupervised Learning in HEG
- Lionel Blonde defends his PhD Thesis in December 2021. His Thesis is entitled "Counterfactual Interactive Learning: Designing Proactive Artificial Agents that Learn from the Mistakes of other Decision Makers".
- Nikolaos Kokkinis-Ntrenis and Marc Desaules have completed their Master Theses while working at the group.
In terms of publications, the team members produced 10 papers in 2021:
- Firstly, Joao's paper 'Conditional neural relational inference for interacting systems' was awarded the Best Student Data Mining Paper Award in the in ECML/PKDD 2021 conference. Congrats Joao!
- Naoya has two accepted papers in NeurIPS 2021, in a research line that seems to be bringing a series of upcoming works related to the integration of physics in machine learning models. The two works are: 1) "Physics-Integrated Variational Autoencoders for Robust and Interpretable Generative Modeling" and 2) "Variational Autoencoder with Differentiable Physics Engine for Human Gait Analysis and Synthesis"
- Jason's work "Kanerva++: extending The Kanerva Machine with differentiable, locally block allocated latent memory" was presented in ICLR 2021.
- Greg's work "Analysing the data-driven approach of dynamically estimating positioning accuracy" was presented in IEEE ICC 2021 conference, while his works "ProxyFAUG: Proximity-based Fingerprint Augmentation" and "Towards Reproducible Indoor Positioning Research" were presented in IPIN 2021.
- Yoann published a preprint of his work "Permutation Equivariant Generative Adversarial Networks for Graphs".
- The work of Magda on learned compression was summarised in the preprint "Learned transform compression with optimized entropy encoding".
- Lastly, Lionel published his pharaonic work: "Where is the Grass Greener? Revisiting Generalized Policy Iteration for Offline Reinforcement Learning"
Research projects are a core part of the group's activities. In the IAI Innosuisse project, the focus is on Industrial Artificial Intelligence for intelligent machines and manufacturing digitalization, In this project, our group works in close collaboration with ABB. Moreover, the analysis of human gait within the SimGait project has been a central theme for our group. Lastly the project Eratosthenes project, funded under the Spark funding scheme the Swiss National Science Foundation (SNSF) focuses on mitigating the barrier of the data collection requirement of fingerprinting localization, through data augmentation and synthetization. You may find more information about projects and partners in the Collaborations page.
This year, we didn't have the chance of organizing a lot of events, but we took the opportunity to meet and share some nice wine on the good-buy party of Magda Gregorova, who took a Professorship in Germany. We hope to be able to organize more meetups in 2022!
The dmml group wishes you happy holidays and a happy 2022!!!