
Ph.D. candidate in Ingegneria Elettrica, Elettronica E Delle Comunicazioni , 39th cycle (2023-2026)
Department of Electronics and Telecommunications (DET)
Profile
PhD
Research topic
The thesis aims at developing techniques for deep learning-based multi-modal image processing
Tutors
Research interests
Biography
Luca Dordoni is a doctoral student at the DET, part of a research group aiming to utilize deep learning techniques for image processing.
He completed a Bachelor's degree in Physical Engineering at Politecnico di Torino, which led him to pursue a Master's degree in Physics of Complex Systems at the same institution. It was during this journey that he became acquainted with the world of deep learning, which fascinated him. Consequently, he embarked on a research study for his master's thesis on resource optimization techniques used in artificial neural networks titled "Sparsification of Deep Neural Networks via Ternary Quantization". Through this work, he developed a ternary quantization method based on quantization-aware training for the development of lightweight deep neural networks, essential in modern technologies for deployment on embedded devices or those with limited memory capacity.
His current doctoral research primarily revolves around image processing, with emphasis on multi-modal learning, specifically on depth super-resolution tasks, which involve enhancing the resolution and quality of depth maps obtained from various sensors, such as LiDAR or stereo cameras. He is particularly interested in finding efficient architectures achieved through the application of quantization techniques. These methods aim to streamline the computational resources required for image processing tasks, such as depth super-resolution, while maintaining high performance standards across various modalities.
He completed a Bachelor's degree in Physical Engineering at Politecnico di Torino, which led him to pursue a Master's degree in Physics of Complex Systems at the same institution. It was during this journey that he became acquainted with the world of deep learning, which fascinated him. Consequently, he embarked on a research study for his master's thesis on resource optimization techniques used in artificial neural networks titled "Sparsification of Deep Neural Networks via Ternary Quantization". Through this work, he developed a ternary quantization method based on quantization-aware training for the development of lightweight deep neural networks, essential in modern technologies for deployment on embedded devices or those with limited memory capacity.
His current doctoral research primarily revolves around image processing, with emphasis on multi-modal learning, specifically on depth super-resolution tasks, which involve enhancing the resolution and quality of depth maps obtained from various sensors, such as LiDAR or stereo cameras. He is particularly interested in finding efficient architectures achieved through the application of quantization techniques. These methods aim to streamline the computational resources required for image processing tasks, such as depth super-resolution, while maintaining high performance standards across various modalities.
Publications
Latest publications View all publications in Porto@Iris
- Dordoni, Luca; Migliorati, Andrea; Fracastoro, Giulia; Fosson, Sophie; Bianchi, Tiziano; ... (2024)
Sparsification of Deep Neural Networks via Ternary Quantization. In: 34th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2024, London (UK), 2024, pp. 1-6. ISSN 2161-0371. ISBN: 9798350372250
Contributo in Atti di Convegno (Proceeding)