Research
These are the main topics currently developed by NESYA in publications, projects and students' theses.
Other topics that fall within the various application fields of NESYA are also possible of course, please contact us!
Artificial Intelligence for Green and Sustainable Technologies
Energy management is a key factor for the growth and development of the society. Nowadays, the forecasting of energy consumption and production has become a crucial need to improve energy performance and sustainability. In the framework of renewable energy sources, the development of novel deep learning paradigms represents an important challenge for sustainable development. We are applying several different technologies and studying novel algorithmic frameworks for realizing synergy among different solutions in the "Twin Transition" of energy and digital systems. Some possible B.Sc. and M.Sc. theses may deal with:
energy time series forecasting (load and production), especially for renewable energy sources;
advanced optimization models based on neural networks and generative techniques, for application in the green and smart energy sectors;
machine learning and deep learning applied to smart grids, smart buildings, energy communities, smart sensors networks, etc.
Hyperdimensional Computing & Vector Symbolic Architectures
Application of HDC/VSA computational frameworks, which are based on random distributed representations, to various stages of neural network design and application. In particular, we are studying: HD vectors as input/output to neural networks and their application in randomized learning, HDC/VSA primitives in neural network design, explainability and interpretability of neural networks using HDC/VSA. The long-term goal is to study the possibility of a whole artificial intelligence framework based on these concepts, leveraging on brain-inspired and brain-like technologies. Some possible M.Sc. theses, also in collaboration with the University of California Berkeley and the Luleå University of Technology (Sweden), may deal with HDC/VSA applied to computer vision, wireless communication, language processing, classification, time series prediction and neural modeling.
Quantum AI: Circuits, Algorithms and Applications
The massive amount of data produced and the increasing complexity of traditional deep learning models requires the development of new scalable systems and algorithms. A novel area of research is represented by Quantum Artificial Intelligence (QAI), which leverages the laws of quantum mechanics to efficiently deal with large-scale heterogeneous data. A hybrid quantum-classical approach based on Variational Quantum Circuits (VQC) represents the most promising solution to exploit near-term devices and benefit from the properties of quantum technologies. We focus on the latest applications of quantum machine learning and quantum deep learning on near-term devices, as well as on its possible quantum advantage. The variational learning framework is analysed to tackle classically hard-to-solve problems, such as portfolio optimization, high-dimensional data classification and simulation of high energy physics events. Some possible M.Sc. theses, also in collaboration with CERN with Φ-lab of European Spatial Agency (ESA), may deal with:
design of Quantum Convolutional, Recurrent or Graph Neural Networks through VQC able to run on near-term quantum devices for efficient data processing;
Quantum Convolutional Neural Networks for massive image processing and analysis in the Multispectral Satellite Imagery and High Energy Physics (HEP) domains;
Quantum Generative Adversarial Networks for massive generation of synthetic data in the HEP domain;
analysis of the variational learning framework, exploring the possibilities of theorizing new ansatz models, addressing the barren plateaus problem, and encoding data efficiently with parameterized quantum circuits;
investigation of Quantum Neural Networks’ capacity, expressivity and generalization abilities over real-world data and applications.
Quantum Optimization Algorithms
Quantum optimization in the context of quantum computing leverages the principles of quantum mechanics to enhance the efficiency of machine learning algorithms. The main objective is to find optimal or approximate solutions to optimization problems, which are crucial in many machine learning applications such as model training. Unlike classical optimization algorithms, quantum algorithms can harness the unique properties of superposition and entanglement to achieve advantageous properties, such as lower error rates or faster convergence rates. In this framework, potential topics of M.Sc. theses could include the implementation of new quantum optimization algorithms, the application of well-known quantum optimization algorithms in contexts of high practical interest, or the integration of classical machine learning systems with quantum optimization procedures using a hybrid approach in order to systematically leverage the advantages of both classical and quantum approaches.
Emotion Recognition and Behavioral Analysis
Emotion recognition is the process of identifying the human emotions and it consists in a classification system that processes unstructured data, revealing the mood of the analyzed people. Human emotion can be recognized from facial expressions, speech, gestures, posture, or even by psychological signals. Scientific achievements obtained in this topic can be applied to a large extent of practical real-world applications such as e-learning, recommendation systems, health monitoring, customer profiling, satisfaction related systems, and so on. Some possible M.Sc. theses, also in collaboration with Machine Learning Solutions, may deal with:
multimodal analysis of emotion recognition by using machine learning applied to several data streams related to different information channels as videos, images, speech, text, etc.;
biomedical applications with the opportunity to develop deep learning models for healthcare monitoring and telemedicine, from illness detection to disease predisposition analysis.
Supervised and Unsupervised Federated Learning
By using federated learning it is possible to train an algorithm across multiple edge devices working on local datasets, without exchanging data themselves. In addition, decentralized (or distributed) approaches assume that local datasets are identically distributed. Federated learning allows us to build machine learning models without sharing data, thus allowing to address critical issues such as data privacy, data security, data access rights and access to heterogeneous data. Some possible B.Sc. and M.Sc. theses may deal with:
edge machine learning and IoT applications;
distributed algorithms for clustering, classification, data regression and time series prediction;
federated learning applied to logistics, finance, e-Health, smart grids, e-Mobility, etc.
Weareable Devices for Telemedicine and Biomedical Applications
With the advent of new wearable devices and smartwatches, a new problem is attracting the interest of industries and research centers worldwide. Namely, it is necessary to study the development, adaptation and assessment of Artificial Intelligence (AI) and machine learning algorithms in order to exploit the future availability of big data related to health applications. Some possible M.Sc. theses, also in collaboration with ESSEDH, may deal with:
Internet of Medical Things (IoMT) device classification procedures;
study of digital clinical measurements with data driven models and AI-based algorithms for biomarkers evaluation;
benchmarking methodologies of the devices according to the digital clinical data collected by them.
Artificial Intelligence for Smart Tourism and Cultural Heritage
This research activity aims at developing new data mining and data retrieval systems based on intelligent algorithms that exploit machine learning and deep learning techniques for the use of dynamic and personalized contents on the portable tourist's devices. Some possible M.Sc. theses, also in collaboration with Manet Mobile Solutions, may deal with:
state-of-the-art methodologies for pattern analysis using machine learning and deep learning;
extraction of tourist contents and context-aware computing;
clustering of tourist sites and analysis of tourist paths;
time series analysis for tourist recommendation systems and site rating;
development of intelligent algorithms for analyzing the tourist's behavior on a geographical basis.
