Dilated convolution with learnable spacings

I would like to express my deep gratitude to the institutions and people who made my Ph.D. possible. First and foremost, I would like to thank the French National Research Agency (ANR) and the Artificial and Natural Intelligence Toulouse Institute (ANR-3IA ANITI), as well as the Toulouse Occitanie region, for generously funding my Ph.D. studies. I am also grateful to the "AI for physical models with geometric tools" chair, headed by Prof. Fabrice Gamboa, for their continuous support. My work has greatly benefited from the HPC resources of IDRIS and GENCI's Jean Zay computing center (grants 2021-[AD011013219] and 2023-[AD011013219R1]) and CALMIP computing center (grants 2021-[P21052] and 2023-[P22021]). I sincerely appreciate the contributions of all the staff at both computing centers for enabling us researchers and graduate students to achieve our research goals. Heartfelt gratitude is owed to my thesis supervisor, Timothée Masquelier, for believing in me and proposing this thesis topic, on which we worked together for a little more than 3 years. We have always worked in an atmosphere of trust and scientific exchange. I particularly appreciated the fact that Timothée's office was always open to me. Whatever ideas I had, good or bad, Tim was there to hear them and to suggest improvements or contradict me if they weren't relevant. I particularly appreciated Timothée's ability to adapt his guidance to my needs. I never felt belittled, but always challenged and supported, and I sincerely thank him for that. My sincere appreciation goes to my thesis co-supervisor Thomas Pellegrini for his mentorship, for our weekly exchanges, and for being there when I really needed him! I would like to extend my deepest thanks to the jury of my PhD thesis for their invaluable time, insightful questions, and constructive comments. In particular, I express my gratitude to the rapporteurs of my thesis, Prof. Nicolas Thome and Prof. Emre Neftci, for their thorough examination and valuable feedback. I would like to thank my co-authors: Ilyass Hammouamri from the NeuroAI team at CerCo and Etienne Labbé from the SAMOVA team at IRIT. vi I would also like to thank Wei Fang, although I didn't work with him directly, for developing the SpikingJelly framework that we used in the fourth chapter of my thesis. I want to express my thanks to the research teams in which I have worked over the last few years, in particular the NeuroAI team at CerCo. I would especially like to thank the permanent members of the team, Timothée Masquelier, Rufin VanRullen for his advice and wise remarks, and Andrea Alamia, whom I knew as a post-doc, then as a researcher at the CNRS and finally as an HDR researcher. Many thanks to all those with whom I shared a workspace, especially my office colleagues, for their patience and understanding during our sometimes lively exchanges.