quentindelfosse.me quentin (dot) delfosse (at) cs (dot) tu-darmstadt (dot) de
Meetings by appointment.
Mission. My research focuses on how Explainable Reinforcement Learning. I try to understand how knowledge is represented by agents that need to model and interact with their environments. I am also a 3D printing enthusiast. Please find more info here.
|2020 - now:||Ph.D. student at the Machine Learning Lab, CS Department, TU Darmstadt, Germany|
|2017 - 2019:||M.Sc. MoSIG in Computer Science at UGA Grenoble (Double Diploma), France|
|2016 - 2019:||Dipl. Ing. in Computer Science at ENSIMAG Grenoble, France|
|2014 - 2016:||Higher School Preparatory Class at Lycée Buffon Paris, France|
Rational Activation Functions
Rationals as learnable activation functions, trainable in an end-to-end fashion using backpropagation.
Neural Plasticity for Deep RL
Allowing learning structure to embed high plasticity (with rationals) is crucial for dynamic RL environments.
Object Detection with Time Consistency
Using space-time consistency to better detect and understand objects.
|2021||Siyao Chen, Comic Generation with Rational Generation Model, M.Sc. Thesis, co-supervision Arseny Skryagin|
|2021||Maximilian Otte, Creating Emojis with Generative Adversarial Neural Cellular Automata, M.Sc. Thesis, co-supervision Johannes Czech|
|2021||Julius Zimmermann, Evaluation and Showcasing of Rational Functions, M.Sc. Thesis, co-supervision Patrick Schramowski|
|2020||Patrick Vimr, Optimizing the analysis of Rational Networks applied to Deep Reinforcement Learning-Algorithms, B.Sc. Thesis|
|WS 2020/21||Quentin Delfosse, Patrick Schramowski, Improving the Rational Activation Functions library, Bachelor Praktikum|