Wednesday, April 27, 3.15 pm, 2022

Recent advances in black-box models for continuous-time dynamical systems
Çağatay Yıldız, Department of Computer Science, Aalto University.
Abstract: Since the neural ordinary differential equations (ODE) breakthrough, continuous-time dynamical systems have gained enormous popularity. In this talk, we will present a summary of recent advances in black-box continuous-time systems. We will start with the a gentle introduction to ODEs, followed by Gaussian process based, uncertainty-aware ODE systems. Afterwards, we will discuss how to learn stochastic systems using a similar framework. We then explain ODE2VAE, a deep generative second-order ODE model for learning video sequences. Finally, a novel continuous-time model-based reinforcement learning approach will be presented. The presentation will be based on the dissertation here.
Location: Zoom
Organizer: Fredrik Lindsten

Wednesday, May 4, 3.15 pm, 2022

Learning Proposals for Probabilistic Programs with Inference Combinators
Heiko Zimmermann, Amsterdam Machine Learning Lab (AMLab), University of Amsterdam.
Abstract: We develop operators for construction of proposals in probabilistic programs, which we refer to as inference combinators. Inference combinators define a grammar over importance samplers that compose primitive operations such as application of a transition kernel and importance resampling. Proposals in these samplers can be parameterized using neural networks, which in turn can be trained by optimizing variational objectives. The result is a framework for user-programmable variational methods that are correct by construction and can be tailored to specific models. We demonstrate the flexibility of this framework by implementing advanced variational methods based on amortized Gibbs sampling and annealing.

Paper: https://arxiv.org/pdf/2103.00668.pdf
Location: Alan Turing
Organizer: Fredrik Lindsten