Capsule networks: structure and function for digit recognition
7 luglio alle 11,00. presso la sala riunione 8° piano in Via Celoria 18
Speaker: Barışcan Kurtkaya
Responsabile: N. Alberto Borghese
Capsule networks are a novel deep neural networks architecture that is promising in terms of explainability and potentiality. Their structure is reviewed and compared to classical Convolutional Neural Networks to highlight similarities and differences. Issues related to kernel configuration and results on handwritted digit recognition are reported and discussed.
Barışcan Kurtkaya is a 4th-year student in Electronics and Communication Engineeringat Yıldız Technical University, Turkey and he is currently an Erasmus student at Università degli Studi di Milano in the laboratory of Applied Intelligent Systems. His research interests are in deep learning architectures, 3D computer graphics, computer vision, robotics, and NLP and has carried out research projects on semi-autonomous underwater vehicles, driver assistance systems with computer vision, and unmasked person detection systems with deep learning on which I am completing a journal paper. Recently I got involved on capsule networks as an instruments to provide more robust learning results.