Rembert Daems successfully defended his PhD thesis, which explores how machines can learn to understand the world from video alone.

His research bridges deep learning, stochastic processes, and physics-informed modeling to address a big challenge in robotics and AI: learning dynamics from high-dimensional visual input.

Rembert developed new methods to learn physical and stochastic models directly from image sequences, enabling more efficient prediction and control in robotic settings. His contributions include a framework for learning Lagrangian dynamics from pixels, a novel approach to variational inference in neural stochastic differential equations driven by fractional Brownian motion, and a control-inspired strategy for improving the efficiency of training such models. Congratulations to Rembert!