Towards Learning Controllable Representations of Physical Systems

no code implementations • 16 Nov 2020 • Kevin Haninger, Raul Vicente Garcia, Joerg Krueger

Learned representations of dynamical systems reduce dimensionality, potentially supporting downstream reinforcement learning (RL).

reinforcement-learning Reinforcement Learning (RL)

Uncertainty Quantification in Deep Residual Neural Networks

no code implementations • 9 Jul 2020 • Lukasz Wandzik, Raul Vicente Garcia, Jörg Krüger

In this work, we address the problem of uncertainty quantification in deep residual networks by using a regularization technique called stochastic depth.

Uncertainty Quantification