SEED Research topics

Edge-TinyML Visual Plant Disease Detection for Next-Generation Precision Farming

ID: SEED-PhD-1-precision-farming PDF version

This doctoral research proposal presents a novel solution through energy-efficient Edge-TinyML sensor implementation that integrates (i) backscatter wireless information and power transfer (i.e., sensor powered by energy scavenged from radio frequency signals) (WIPT) technology, (ii) efficient WIPT antenna design, and (iii) a low-cost, efficient processor for TinyML applications.

Leveraging LLM for a full fledged Intent-based Security

ID: SEED-PhD-2-intent-security PDF version

The PhD is expected to improve intent-based networking to security from multiple points of view: 1) definition of new intent specification languages and formats, compliant with security requirements that a network administrator would need to express; 2) investigation of novel AI-empowered algorithms for the translation of natural languages to the newly defined intent specification languages; 3) study of how to provide assurance for the usage of LLMs for the translation from natural language to intent languages; 4) investigation of algorithms for intent translation into policies

Low-Resource Music Synthesis and Editing for Video Games

ID: SEED-PhD-3-music-synthesis PDF version

Generative and creative systems based on Deep Learning have recently emerged under the umbrella term “Generative Artificial Intelligence” (Gen-AI), and are being quickly adopted by the general public, the most notable examples being conversational models such as ChatGPT, text-to-image models such as Midjourney, and text-to-music systems like YuE. However, most Gen-AI systems have been trained using large scale data scraping (e.g. books, articles, photos, art pieces or commercial music hosted on publicly accessible platforms) and potentially breach copyrights. The goal of this PhD topic is to investigate low-resource music synthesis and editing for video games.

Detecting anomalies in temporal networks, providing explanations and supporting multi-criteria decision in defence applications

ID: SEED-PhD-4-anomaly-detection PDF version

This thesis focuses on anomaly detection, understanding, and labeling in complex networks for socially impactful applications such as social networks, financial exchange, health, defence, energy, etc. The two main challenges are: (i) the limited access to labeled data for anomaly detection, (ii) and when labels are obtained, they are often incorrect or unusable due to errors made by domain experts in labeling anomalies. To address these challenges, we propose to take advantage of three research areas: anomaly detection (for graphs), explainable AI (XAI), multi- criteria decision aiding (MCDA).

Deep learning-aided error-correction for storage on synthetic DNA molecules

ID: SEED-PhD-5-storage-dna PDF version

DNA is a robust medium capable of withstanding significant temperature variations and it is durable over time. It is therefore expected to significantly reduce the energy consumption of data storage. Research on this area is multidisciplinary by nature, encompassing fields as diverse as biology, chemistry, bioinformatics, signal processing, and error-correction. This PhD falls into the fields of error-correction and Deep Learning. Due to the inherent unreliability of the DNA storage support, the goal will be to develop advanced deep learning models to ensure perfect data recovery after storage.

PeRSAFE: Personalizing Reverse Shoulder Arthroplasty planning with Finite Elements

ID: SEED-PhD-6-shoulder-arthroplasty PDF version

Finite element (FE) modelling is an engineering tool for structural analysis that has been used for many years to assess the relationship between load transfer and bone morphology. Patient-specific FE models have shown to be valuable tools to optimise the design and fixation of orthopaedic implants, to assess the risks of fractures or to optimise pre-operative planning of implant placement. However, FE models are limited by their stability, generalisation ability, and computational speed thus restricting their wide adoption in clinical routine. The PhD objectives are therefore: (1) to quantify the variability of subject-specific material properties of the scapula in RTSA patients, to further define boundary and loading conditions based on individualised data; (2) to speed up FE model computation through machine learning prediction, in order to make it usable in clinical routine; (3) to conduct experimental validation of FE prediction results, in order to establish their significance for clinical improvements.

Performance metrics of building ventilation conditions and mitigation actions to improve indoor air quality and well-being

ID: SEED-PhD-7-air-quality PDF version

The proposed thesis focuses on the components of the TAIL certification rating scheme, specifically the Indoor Air Quality (IAQ) component. The questions to be addressed are: 1) Can ventilation and remediation scenarios be recommended according to building typology and use? 2) Can IAQ and IEQ improvement be demonstrated with a global rating scheme like the TAIL rating system under actual conditions? 3) How can the evaluation system evolve for large-scale use? Through the process, it is intended to further advance the TAIL scheme by extending for human responses and, if needed, adding new elements and making compact elements that exist already, such as parameters defining IAQ.

Analysis of uncertainty propagation and decision-making effects associated with the optimal design and operation of multi-energy systems

ID: SEED-PhD-8-energy-systems PDF version

While existing methods tackle these issues separately, our goal is to develop integrated policies that leverage auxiliary technologies, storage, suppliers and external customers to improve the resilience of multi-energy networks design and operations flexibility. This can increase economic, environmental and even social impacts, influencing decision-makers and stakeholders’ interest.

Data Sharing Motivations and Practices in Innovation Districts

ID: SEED-PhD-9-open-collaboration PDF version

In the context of a world that is increasingly producing more data that has potential to drive better and more evidence-based decision making, the motivations and incentives to acquire, share, structure, and model data are inconsistent. The consequence of which is that alignment around goals is difficult and the coordination of actors to increase the benefits of governmental investment is inherently problematic. Better understanding of the interest of each stakeholder to switch to an open data organisation, as the best socio-technical systems for open collaboration, remains to be understood. This multidisciplinary research aims to address this challenge.

Efficient foundation models for ocean remote sensing observations

ID: SEED-PhD-10-ocean-observation PDF version

Recent methods based on state-space models [3] have demonstrated strong capabilities in modeling very long sequences. In this context, these methods provide the perfect alternative to standard deep learning approaches in representing long correlation in physical observations. This PhD aims to leverage these state-space representations to develop foundation models of ocean satellite data observations. These models are expected to better capture long-range dependencies more effectively and to improve the detection and classification of physical phenomena.

Image Reconstruction for Low field MRI

ID: SEED-PhD-11-image-reconstruction PDF version

The scientific objective of this PhD will focus mainly on advanced AI methodologies. Incorporating physics-guided deep learning models that explicitly integrate the underlying MRI signal formation process to enhance reconstruction reliability and interpretability. To this end, part of the project will be dedicated to the development of efficient computational strategies: Achieving real-time image reconstruction necessitates optimized numerical solvers and meta-learning techniques for rapid inference at the point of care.

Fine-grained and customized, neuromusculoskeletal-based assessment of rehabilitation exercises

ID: SEED-PhD-12-rehabilitation-exercise PDF version

This PhD will pursue interdisciplinary research by accounting both for visual features as well as latent neuromusculoskeletal features of body movement. To account for the latter, visually captured human movement will be further instantiated via biomechanical digital twins (DT) of the human through tools used in the medical domain, namely, OpenSIM and Hufydy that account for the biomechanical structure of the human body as well as physics constraints.

Energy-aware actor-based distributed programming

ID: SEED-PhD-14-energy-programming PDF version

The main goal of this PhD is the precise definition and efficient enforcement of energy contracts over distributed programs. We are targeting a distributed programming language extension that allows required and available energy quotas of programs to be partially defined by the programmer and partially provided dynamically by the distributed environment in which the program is executed.

Acceleration and topology reduction of power flow simulations via machine learning

ID: SEED-PhD-15-power-flow PDF version

This PhD has two main objectives: 1. Creating a scalable hybrid model for simulating large electricity networks with fixed topologies, supporting applications like real-time state estimation, impact assessment of emerging electricity uses (e.g., EVs, heat pumps), and uncertainty management in power flow calculations. 2. Extending this model to handle variable topologies, transforming it into a machine-learning-accelerated power flow solver.

New detectors embedded in liquid xenon for next generation high-performance telescopes

ID: SEED-PhD-16-photodetector-telescope PDF version

This PhD proposes a twofold joint use of the XEMIS1 setup for the following purposes: 1. R&D for the next generation LXe detectors; 2. open the French-Australian collaboration towards medical imaging. Point 1) foresee the use of Compton liquid Xenon to test several models of PMTs and SiPM that might be interesting for a future generation LXe observatory. Point 2) will allow to start new joint research working relationships, that may include also industrial partners in the future, through the co-development of new advanced detectors or detector components.

Flow Management Optimization in SCHC compression framework

ID: SEED-PhD-17-flow-management PDF version

This PhD focuses on extending SCHC's capabilities to manage diverse flow types, particularly TCP and QUIC protocols, with potential applications to RTP streams. The primary research objective involves developing anticipatory mechanisms for protocol header evolution to maintain optimal compression efficiency, even under challenging network conditions characterized by packet loss and extended Round-Trip Times (RTT). A significant application domain for this research lies in interplanetary communication.

Generative AI for IoT Interoperability: Enhancing Schema Translation and Extending SID Registry Applications

ID: SEED-PhD-18-iot-interoperability PDF version

This PhD addresses the challenges of IoT interoperability by exploring the integration of generative AI with a SID registry to create a universal schema translation and management framework. Three key challenges are identified: automated schema translation, SID registry enhancement and validation and performance optimization.

Modeling of PFAS thermal destruction in Hazardous Waste incinerator

ID: SEED-PhD-19-pfas-incineration PDF version

The aim of this thesis is to model the incineration of PFAS laden hazardous waste on an industrial scale. An exhaustive bibliography on the destruction of PFAS in hazardous waste incinerators will take into account the oxygen content of the gas phase.

Exploring Realistic Skin Numerical Models in THz Spectroscopy

ID: SEED-PhD-20-skin-models PDF version

This project aims to develop an EM solver for highly realistic skin models. The simulation tool will combine a full-wave skin model together with a physical optics-based model of the spectrometer. By hybridizing a regularized full-wave volume integral equation (VIE) with a physical optics-based solver (HF), we will tackle challenges such as the spectrometer's large electrical size, skin anisotropy, and the coupling between the sample and the spectrometer

Integration of retinal imaging optics in an instrumented contact lens

ID: SEED-PhD-21-instrumented-lenses PDF version

Design and manufacture instrumented contact lenses for healthcare and professional applications linked to augmented reality. A significant part of the PhD will involve building a complete model of the lens associated with the eye, in order to dimension the components and study and optimize the tolerancing of the assembly before manufacturing a prototype. The student will contribute to the development of a phantom eye and a radiometric measurement bench integrating the constraints of retinal imaging.

Optimization of complex multi-carrier energy systems guided by Machine Learning Algorithms

ID: SEED-PhD-22-multi-energy PDF version

The design and operation of today's energy systems requires the inclusion of multiple technologies and resources, including external energy supplies and primary sources. In this PhD machine learning algorithms are used to generate models that allow identifying automatic solution strategies for the optimization of energy systems.