âSignal Processing in the AI eraâ was the tagline of this yearâs IEEE International Conference on Acoustics, Speech and Signal Processing, taking place in Rhodes, Greece.
In this context, Brent de Weerdt, Xiangyu Yang, Boris Joukovsky, Alex Stergiou and Nikos Deligiannis presented ETRO’s research during poster sessions and oral presentations, with novel ways to process and understand graph, video, and audio data. Nikos Deligiannis chaired a session on Graph Deep Learning, attended the IEEE T-IP Editorial Board Meeting, and had the opportunity to meet with collaborators from the VUB-Duke-Ugent-UCL joint lab.
Featured articles:
- De Weerdt, B. et al., âDesigning Transformer Networks for Sparse Recovery of Sequential Data Using Deep Unfoldingâ (https://ieeexplore.ieee.org/abstract/document/10094712)
- Yang X. et al., âRelevance Propagation through Deep Conditional Random Fieldsâ (https://ieeexplore.ieee.org/abstract/document/10095075)
- Stergiou A. et al., âPlay it back: Iterative attention for audio recognitionâ (https://ieeexplore.ieee.org/abstract/document/10096532, developed at the University of Bristol)
On November 7th 2024 at 16:00, Boris Joukovsky will defend their PhD entitled â SIGNAL PROCESSING MEETS DEEP LEARNING: INTERPRETABLE AND EXPLAINABLE NEURAL NETWORKS FOR VIDEO ANALYSIS, SEQUENCE MODELING AND COMPRESSIONâ.
Everybody is invited to attend the presentation in room I.0.01 or online via this link.
There is growing use of deep learning for solving signal processing tasks, and deep neural networks (DNNs) often outperform traditional methods little domain knowledge needed. However, DNNs behave as black boxes, making it difficult to understand their decisions. The empirical approaches to design DNNs often lack theoretical guarantees and create high computational requirements, which poses risks for applications requiring trustworthy artificial intelligence (AI). This thesis addresses these issues, focusing on video processing and sequential problems across three domains: (1) efficient, model-based DNN designs, (2) generalization analysis and information-theory-driven learning, and (3) post-hoc explainability.
The first contributions consist of new deep learning models for successive frame reconstruction, foreground-background separation, and moving object detection in video. These models are based on the deep unfolding method, a hybrid approach that combines deep learning with optimization techniques, leveraging low-complexity prior knowledge of the data. The resulting networks require fewer parameters than standard DNNs. They outperform DNNs of comparable size, large semantic-based convolutional networks, as well the underlying non-learned optimization methods.
The second area focuses on the theoretical generalization of deep unfolding models. The generalization error of reweighted-RNN (the model that performs video reconstruction) is characterized using Rademacher complexity analysis. This is a first-of-its-kind result that bridges machine learning theory with deep unfolding RNNs.
Another contribution in this area aims to learn optimally compressed, quality-scalable representations of distributed signals: a scheme traditionally known as Wyner-Ziv coding (WZC). The proposed method shows that deep models can retrieve layered binning solutions akin to optimal WZC, which is promising to learn constructive coding schemes for distributed applications.
The third area introduces InteractionLIME, an algorithm to explain how deep models learn multi-view or multi-modal representations. It is the first model-agnostic explanation method design to identify the important feature pairs across inputs that affect the prediction. Experimental results demonstrate its effectiveness on contrastive vision and language models.
In conclusion, this thesis addresses important challenges in making deep learning models more interpretable, efficient, and theoretically grounded, particularly for video processing and sequential data, thereby contributing to the development of more trustworthy AI systems.
Naar aanleiding van de heropening van het Sint-Janshospitaal museum (Museum Sint-Janshospitaal) in Brugge op 16 december zijn de schilderijen van Hans Memling gedigitaliseerd. In dit complex process speelt Universum Digitalis, een spin-off van het Departement Elektronica en Informatica (ETRO) van de VUB, een cruciale rol. De spin-off, onder leiding van professor Frederik Temmermans, verwerkte in opdracht van Musea Brugge honderden hoge resolutie detailopnames die door het Koninklijk Instituut voor het Kunstpatrimonium (KIK-IRPA) met nauwgezette methodiek werden gecapteerd. Universum Digitalis stoomde giga-pixel beelden klaar voor gebruik op het web, maar ook voor verder onderzoek in allerlei toepassingen.
De werken zijn online beschikbaar in hoge resolutie (closertomemling.be), en laten bezoekers toe om de kleinste details van de meester vele malen uitvergroot te bewonderen. Begin 2024 volgen ook andere belichtingstechnieken zoals infrarood, streeflicht en x-stralen.
Dit is niet het eerste wapenfeit van Universum Digitalis. De spin-off spitst zich toe op ultra hoge resolutiebeeldverwerking bij digitalisering van het onschatbare kunstpatrimonium, maar ook de verwezenlijking van interfaces die dit soort beelden eenvoudig toegankelijk maakt voor zowel de kunstwetenschapper als de museumbezoeker. Ze bouwden toepassingen voor musea, culturele instellingen en anderen voor de digitale reproductie van schilderijen/kunst in extreem hoge resolutie: van webtoepassingen tot museumkiosken. Voorbeelden zijn de werken van Bruegel (insidebruegel.net) en het oevre van Van Eyck (closertovaneyck.kikirpa.be), een project dat eerder bekroond werd met een Belgian E-Gov Award en European Heritage / Europa Nostra Award.
Contact:
Frederik Temmermans: : frederik.temmermans@vub.be 0479 49 94 26
His speach is entitled “Klimaatverandering in een historisch, mondiaal en antropogeen perspectief“, or “climate change in a historical, mondial and antropogene perspective”. In this short introduction, we look at the current climate change at a larger time scale and across the different regions of the world. We highlight some of the relationships with the world population increase and the local and mondial economic activity, focusing on energy consumption. We finish the talk with a vision on the future in the domains of energy sources and transitions in automotive
More info at:
- First PhD in collaboration with Technical University Twente about ISFETs. (Luc Bousse).
- Oscar Steenhaut – Rector at VUB.
Tom Godden, Thibaut Vandervelden and Ruben De Smet (ETRO) participated to the FENTEC Crypto Hackathon This is an end project of a H2020 from ao. KUL en Atos. The team ended on the second place and won 2000âŹ.
- If it is the intention, to apply for a FWO travel grant, then DO NOT use the VUB OMNIA travel platform to book your trip!
Otherwise, if FWO approves your travel grant, it wonât be possible anymore to get reimbursement for your travel, because if booked via the VUB travel platform, OMNIA will send an invoice to VUB and FWO only pays invoices directly to OMNIA, and addressed to FWO.
OMNIA is not willing to make a credit note for the VUB invoice and charges a fee for re-issuing an invoice.
Here you find the way to book a research travel via OMNIA, if you want to apply for a FWO âTravel Grantâ:
Financien – RESERVERING DIENSTREIS indien gefinancierd met FWO REISKREDIET (service-now.com)
https://vub.service-now.com/sp/?id=kb_article_view&sysparm_article=KB0014417 => English
- OMNIA is one of the VUB official travel agencies! VUB has a framework contract with OMNIA Travel. (Which you better not use in the above mentioned condition!)
Financien – Dienstreizen: voordelen gebruik raamovereenkomst dienstreizen (service-now.com)