DVBF: Deep Variational Bayes Filtering

DVBF is a method, based on variational inference, which can represent time series data in regularised latent spaces.  Our first paper on DVBF describes the theory and its use as deep Kalman filter, creating physically useful latent spaces from high-dimensional observed data.


DVBF can be combined with Dynamic Movement Primitives (DMPs) to obtain robust movement representation. In this paper we explain how not only the variational autoencoder parameters but also the DMP parameters are learned.

Picture of  Justin Bayer

Justin Bayer

Picture of  Maximilian Karl

Maximilian Karl

TUM: PhD candidate
efficient inference
Picture of  Maximilian Soelch

Maximilian Soelch

TUM: PhD candidate
robot control with deep learning
Picture of  Nutan Chen

Nutan Chen

TUM: PhD candidate
hand modelling
Picture of  Patrick van der Smagt

Patrick van der Smagt

current: Head of AI Research, data lab, VW Group

Previous: Director of BRML labs
fortiss, An-Institut der Technischen Universit√§t M√ľnchen
Professor for Biomimetic Robotics and Machine Learning, TUM

Chairman of Assistenzrobotik e.V.

Maximilian Karl, Maximilian Soelch, Justin Bayer, Patrick van der Smagt (2016). Deep Variational Bayes Filters: Unsupervised Learning of State Space Models from Raw Data. arxiv. www bibtex

Nutan Chen, Maximilian Karl, Patrick van der Smagt (2016). Dynamic Movement Primitives in Latent Space of Time-Dependent Variational Autoencoders. pdf bibtex