“Signal Processing in the AI era” was the tagline of this year’s 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., “Designing Transformer Networks for Sparse Recovery of Sequential Data Using Deep Unfolding” (https://ieeexplore.ieee.org/abstract/document/10094712)
- Yang X. et al., “Relevance Propagation through Deep Conditional Random Fields” (https://ieeexplore.ieee.org/abstract/document/10095075)
- Stergiou A. et al., “Play it back: Iterative attention for audio recognition” (https://ieeexplore.ieee.org/abstract/document/10096532, developed at the University of Bristol)
In anticipation of International Women’s Rights Day (March 8), Lesley De Cruz, as one of the 22 researchers gave an interview. Researchers share their daily lives and work as female researchers and scientists, and as women in the field as if you were standing beside them. Discover Lesley’s story, and remember that, more than ever, science is female.
https://www.belspo.be/belspo/scienceconnection/20240308/Lesley_De_Cruz_nl.stm
An ETRO team participated in the 2nd COV19D Competition of the AIMIA Workshop at #ECCV2022 (https://mlearn.lincoln.ac.uk/eccv-2022-ai-mia/). This 2nd COV19D Competition included two Challenges: i) COVID19 Detection and ii) COVID19 Severity Detection. Our team with Abel DĂaz, Tanmoy Mukherjee, MatĂas Bossa, Nikos Deligiannis, Hichem Sahli, and the IT support of Luc van Kempen submitted a solution that beat the Competition’s baseline on both challenges!
The figure illustrates the used method.
The Charcot Fund Jury met on December 9, 2022.
The project “Disentangling cognitive functioning and visual scanning deficits in cognitive test scores” (Prom: Prof J. Van Schependom), has been selected by the Jury for the Charcot Fund 2023.
The Charcot Fund Ceremony will take place on 31 January 2023 at the University Foundation.
On March 29th 2024 at 16::00, Lucas Santana will defend their PhD entitled “TOWARDS UNCHARTED TERRITORIES: HIGH-PERFORMANCE AND HIGH-BANDWIDTH RINGAMP-BASED DELTA-SIGMA ADCs”.
Everybody is invited to attend the presentation in room D.2.01, or digitally via this link.
Analog-to-digital (ADC) research often happens in an agnostic detachment from the intended application; although motivation is sometimes presented, it is not always implemented with the proposed prototype. Advancements in ADC linearity and speed enable applications that were nonexistent before to emerge, such as direct RF conversion and 8k camera recording. Most ADC architectures cover all regions of the performance space, being at the forefront of the state-of-the-art for some areas and not so much for others. This high coverage enables the use of the advantages and peculiarities of different architectures across different applications. One notable architecture that does not perform this is the Discrete Time (DT) Delta-Sigma Modulator (DSM) ADC, in which the published state-of-the-art bandwidth front is limited to 20 MHz. This work investigates this limitation, showing that it can be overcome with high-efficiency ring amplifiers (ringamps) and the correct design process. This work presents a prototype for a single loop 3rd-order DT DSM ADC based on ringamps for the loop filter that could double the bandwidth reached by DT DSM ADC at 47.5 MHz and achieve 67 dB of signal-tonoise and distortion ratio (SNDR) when clocked at 950 MHz. It also shows outstanding figures of merit (FoM): the Schreier FoM, FoMs is 167 dB and the Walden FoM, FoMw is 27 fJ per conversion step. The second prototype used time interleaving to improve the sampling rate and bandwidth further and used a noise-coupled (NC) noiseshaping (NS) SAR quantizer to enable aliased noise suppression. It achieved 1.4 GS/s of sampling rate, a decimated bandwidth of 70 MHz at a peak SNDR of 67 dB, with a power consumption of 32 mW; this translated to a FoMs of 160 dB and a FoMw of 143 fJ/c.s. Both prototypes were the first to pave the way to increase the bandwidth in DT DSM ADC efficiently and can still benefit from recent developments in ringamps and noise-shaping SAR ADCs, leading the architecture to conquer even more space in this uncharted territory.
On June 14 2023 at 11.00, Priscilla Benedetti will defend her PhD entitled “SERVERLESS TECHNOLOGIES AND ARTIFICIAL INTELLIGENCE FOR EDGE SERVICE MANAGEMENT”.
Everybody is invited to attend the presentation at the Aula Magna (Great Hall) of the Department of Engineering, University of Perugia (Address: via Goffredo Duranti 93, Perugia) or online via this link.
.
The emergence of the Internet of Things (IoT) ecosystem has exponentially increased the need for real-time and data-intensive applications. It has shifted the computing load from the centralized cloud to peripheral nodes, hence introducing the adoption of edge computing. In edge computing, services are deployed closer to the users and IoT devices. Edge computing provides computation and storage on geographically distributed nodes, some with limited resources. For this reason, resource efficiency and flexibility is fundamental in edge services: To tackle this challenge, serverless computing can be leveraged. It allows to efficiently deploy containerized applications on resource constrained nodes. It aims at providing the required Quality Of Service (QoS) while limiting resource consumption and allows scaling with traffic volume.
In this context, our work aims at analyzing and developing serverlessbased technologies for edge computing applications. It evaluates the use of Artificial Intelligence, namely Reinforcement Learning (RL) techniques, to optimize the scalability and resource efficiency of serverless frameworks on edge computing clusters. The study will be divided into two main focus areas: Firstly, an experimental analysis of serverless computing for IoT and 5G services is done, considering infrastructures with various features and various open-source software. Secondly, the development and analysis of reinforcement-learning tools to enhance the performance of serverless computing on edge clusters is presented. These tools are evaluated on various IoT-based applications, from simple lightweight webservers to complex stream processing pipelines.
Given the growing traction of serverless computing in both academia and industry, the analysis and tools included in this study will provide important insights on its benefits and drawbacks, while enhancing serverless computing performance for edge services deployment and management.
On September 17 2021 at 16.00 Jakub Ceranka will defend his PhD entitled “Advancements in Whole-Body Multi-Modal MRI: Towards Computer-Aided Diagnosis of Metastatic Bone Disease”.
Everybody is invited to attend the online presentation via  this teams link.
Cancer that begins in an organ, such as the lungs, breast or prostate, and then spreads to the bone or other organs, marks the beginning of metastatic disease. The confident detection of metastatic bone disease and the reliable assessment of the tumour load and treatment response is essential to improve patients’ quality of life and increase life expectancy. Magnetic resonance imaging (MRI) has been successfully used for monitoring of metastatic bone disease. Anatomical whole-body sequences offer excellent resolution and sensitivity for the detection of neoplastic cells within the bone marrow. In combination with spatially prealigned functional diffusion-weighted whole-body MRI and apparent diffusion coefficient maps, it allows for focused, efficient, multi-parametric and holistic evaluation of the total tumour volume, diffusion volume and treatment response assessment. One of the major challenges of radiological reading of whole-body MRI in the clinical routine comes from the large amount of data to be reviewed, making lesion detection and quantification demanding for a radiologist, but also prone to error. Additionally, whole-body MR images are often corrupted with multiple spatial and intensity artifacts, which degrade the performance of medical image processing algorithms.
This PhD thesis proposes number of contributions in the medical image processing domain aiming at improving the quality and extending the usability of whole-body multi-modal MRI in the clinical routine. These include spatial groupwise image registration (to align multiple MRI modalities), multi-atlas segmentation (to define the skeleton region of interest), image standardization (to map MRI intensities into comparable ranges) and a deep learning framework for detection and segmentation of metastatic bone disease, as it is pathology of choice for this work. Combined, proposed contributions provide building blocks for a fully automated computer-aided diagnosis (CAD) system for the detection and segmentation of metastatic bone disease using whole-body multi-modal MRI. Finally, an ablation study describing the impact of different CAD system components on detection and segmentation accuracy is provided.