Over the course of the three-year doctoral period, various topics ranging from machine learning, deep neural networks, data augmentation, virtual and augmented reality, cloud computing and web programming, and finally, edge computing has been deepened and explored. The thesis is organised into four chapters, all of which are closely interlinked but with a different topic underpinning them. Chapter 1 deals with the research strands based on machine learning and describes the studies for the optimisation of convolution operators in the field of GPGPU computing, the techniques tackled to solve emotion recognition (carried out by focusing only on the mouth of the subjects), the identification of proteins from their amino acid chain, and the analysis of images for the recognition of certain skin diseases. Chapter 2 of the thesis focuses on the research we did on the topics of virtual reality and augmented reality, which was a cornerstone of my PhD. We tried to improve the quality of teaching certain subjects, such as mathematics, through the use of virtual reality and augmented reality. For example, we conducted interesting research with schools in Umbria, meeting students and professors, to whom we showed use cases we had realised for the representation of mathematical functions and three-dimensional objects useful for teaching. We then focused on the possibility of recreating three-dimensional objects from real objects, real Digital Twins. This allowed us to start a collaboration with a school of higher education in Umbria, the ITS Umbria Academy, for the realisation of some digital twins of industrial machinery useful for Industry 4.0. The knowledge gained in this field was used in another strand, which focused on the tele-rehabilitation of people with neurological diseases. We then applied the knowledge obtained through the use of virtual reality software to machine learning, with which an alternative technique was developed to the classic data augmentation that is often performed during the learning phase of neural networks. Chapter 3 deals with cloud computing, the main strands of research being the creation of a scalable cloud architecture that exploits the potential of open-source software. Web Apps, web programming and the study of Cloud infrastructures have been fundamental. Applications such as LibreEOL were used to tell and publicise our work, and the fundamental insights into Cloud computing led us to receive an important invitation to the AWS summit, where Amazon Italy itself invited us on stage in Milan to talk about how Open Source technologies, Machine Learning and Virtual Reality have contributed positively to the open source community. The results obtained from the study of Cloud infrastructures have been published in scientific journals and represent a milestone in my doctorate. Chapter 4 of the thesis consists of the results obtained from the profound and efficient collaboration between our research group and Professor Sumi Helal's\footnote{\url{https://www.cise.ufl.edu/helal-abdelsalam-sumi/}} research group in the field of IoT. Our experience with cloud computing was indeed crucial and a prerequisite for understanding the complex topics he suggested. The aim of the collaboration was the use of artificial intelligence techniques, machine learning and artificial neural networks for the optimisation of smart city architectures, which require the following three layers: cloud, edge, and sensors. In fact, it is well known that the cloud has costs that scale according to how it is used. This means that if it is used in a very important way, as in the case of a smart city, then the costs could be unsustainable. In particular, it is possible to optimise the push and pull phases of information between the various layers, for example, by aggregating it or sending it only when strictly necessary based on statistical inference techniques. We are also collaborating to model a dataset, which will be made Open Source via the Kaggle platform, to train neural networks to perform the best actions regarding the management of data traffic coming from homes with IoT sensors inside them that monitor parameters of various kinds (such as temperature, light level humidity, etc.) so as to minimise the amount of data exchanged over the network, minimise the use of the cloud, and maximise the use of edge computing, while still guaranteeing the optimal functioning of the infrastructure in the event that it needs to scale up to a level where it can be operated over a city with tens of thousands of connected homes. This three-year period of work, which has seen the development of four distinct research strands, has also been very fruitful in terms of scientific publications. In fact, this paper will describe and list most of the publications produced by the research group and, in particular, those of which the candidate is the author. The high number of 15 publications as conference proceedings, seven journal publications and two book chapters have been achieved. At the time of writing this paper, two other articles have been submitted to scientific journals, and others are planned for the coming months, a clear sign of how hot and extremely topical these issues have been for the scientific community in these years. The results obtained from the study of cloud infrastructures have been published in scientific journals and represent a milestone in my doctorate.

Innovative approaches to some modern complex problems / Damiano Perri. - (2023).

Innovative approaches to some modern complex problems

Damiano Perri
2023

Abstract

Over the course of the three-year doctoral period, various topics ranging from machine learning, deep neural networks, data augmentation, virtual and augmented reality, cloud computing and web programming, and finally, edge computing has been deepened and explored. The thesis is organised into four chapters, all of which are closely interlinked but with a different topic underpinning them. Chapter 1 deals with the research strands based on machine learning and describes the studies for the optimisation of convolution operators in the field of GPGPU computing, the techniques tackled to solve emotion recognition (carried out by focusing only on the mouth of the subjects), the identification of proteins from their amino acid chain, and the analysis of images for the recognition of certain skin diseases. Chapter 2 of the thesis focuses on the research we did on the topics of virtual reality and augmented reality, which was a cornerstone of my PhD. We tried to improve the quality of teaching certain subjects, such as mathematics, through the use of virtual reality and augmented reality. For example, we conducted interesting research with schools in Umbria, meeting students and professors, to whom we showed use cases we had realised for the representation of mathematical functions and three-dimensional objects useful for teaching. We then focused on the possibility of recreating three-dimensional objects from real objects, real Digital Twins. This allowed us to start a collaboration with a school of higher education in Umbria, the ITS Umbria Academy, for the realisation of some digital twins of industrial machinery useful for Industry 4.0. The knowledge gained in this field was used in another strand, which focused on the tele-rehabilitation of people with neurological diseases. We then applied the knowledge obtained through the use of virtual reality software to machine learning, with which an alternative technique was developed to the classic data augmentation that is often performed during the learning phase of neural networks. Chapter 3 deals with cloud computing, the main strands of research being the creation of a scalable cloud architecture that exploits the potential of open-source software. Web Apps, web programming and the study of Cloud infrastructures have been fundamental. Applications such as LibreEOL were used to tell and publicise our work, and the fundamental insights into Cloud computing led us to receive an important invitation to the AWS summit, where Amazon Italy itself invited us on stage in Milan to talk about how Open Source technologies, Machine Learning and Virtual Reality have contributed positively to the open source community. The results obtained from the study of Cloud infrastructures have been published in scientific journals and represent a milestone in my doctorate. Chapter 4 of the thesis consists of the results obtained from the profound and efficient collaboration between our research group and Professor Sumi Helal's\footnote{\url{https://www.cise.ufl.edu/helal-abdelsalam-sumi/}} research group in the field of IoT. Our experience with cloud computing was indeed crucial and a prerequisite for understanding the complex topics he suggested. The aim of the collaboration was the use of artificial intelligence techniques, machine learning and artificial neural networks for the optimisation of smart city architectures, which require the following three layers: cloud, edge, and sensors. In fact, it is well known that the cloud has costs that scale according to how it is used. This means that if it is used in a very important way, as in the case of a smart city, then the costs could be unsustainable. In particular, it is possible to optimise the push and pull phases of information between the various layers, for example, by aggregating it or sending it only when strictly necessary based on statistical inference techniques. We are also collaborating to model a dataset, which will be made Open Source via the Kaggle platform, to train neural networks to perform the best actions regarding the management of data traffic coming from homes with IoT sensors inside them that monitor parameters of various kinds (such as temperature, light level humidity, etc.) so as to minimise the amount of data exchanged over the network, minimise the use of the cloud, and maximise the use of edge computing, while still guaranteeing the optimal functioning of the infrastructure in the event that it needs to scale up to a level where it can be operated over a city with tens of thousands of connected homes. This three-year period of work, which has seen the development of four distinct research strands, has also been very fruitful in terms of scientific publications. In fact, this paper will describe and list most of the publications produced by the research group and, in particular, those of which the candidate is the author. The high number of 15 publications as conference proceedings, seven journal publications and two book chapters have been achieved. At the time of writing this paper, two other articles have been submitted to scientific journals, and others are planned for the coming months, a clear sign of how hot and extremely topical these issues have been for the scientific community in these years. The results obtained from the study of cloud infrastructures have been published in scientific journals and represent a milestone in my doctorate.
2023
Osvaldo Gervasi, Sumi Helal
ITALIA
Damiano Perri
File in questo prodotto:
File Dimensione Formato  
Perri_Damiano_Thesis_2023.pdf

Open Access dal 06/06/2024

Tipologia: Tesi di dottorato
Licenza: Creative commons
Dimensione 34.58 MB
Formato Adobe PDF
34.58 MB Adobe PDF

I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/1311362
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact