The Uncertainty Propagation in Composite Models workshop is hosted and organised by Siemens AG and takes place in Munich 10-11th of October 2019. Siemens have a long tradition of applying machine learning in application critical scenarios. An important requirement for the deployment of machine learning methods is being able to provide an explanation as to why a specific method works and for this uncertainty is essential.

Bayesian methods provide a principled framework for incorporating prior knowledge with observed data. One of the key benefits of this framework is that we can infer what a specific assumption about the world means in light of observed data. However, when the size of datasets grow and the tasks that we aim to solve increase in complexity, answering the question "why something works" gets more and more challenging. Further, when building hierarchical or composite Bayesian models, computational intractabilities lead to challenges in their interpretation.

This workshop will focus on these aspects of machine learning, specifically: (i) how to propagate uncertainty through composite models, (ii) how to interpret a composite probability distribution and (iii) how to evaluate success. Inspired by reinforcement learning we want to encourage a discussion and debate on what abstract tasks can be constructed that would provide an interpretable measure of success.

Venue

The workshop will be located in Central Munich at Hotel Louis, Viktualienmarkt 6, 80331 München. Right next to the hotel is Siemens AI-Lab where we will have a welcome breakfast and introductions on Thursday morning. Instructions on how to get there can be found here.

Schedule

The schedule is still tentative as we are waiting for a few more confirmations. As soon as abstracts and titles starts trickling in we will put them here.

Wednesday 9th of October

As we will start early on Thursday morning most of us will arrive on Wednesday evening. Because it is getting a bit cold for Munich's famous beer gardens, we will do the next best thing and visit one of the traditional breweries in the city centre, right next to Viktualienmarkt. We will meet at Paulaner im Tal https://www.paulaner-im-tal.de/ at around 18:30. Feel free to join us for food or a drink.

Thursday 10th of October

Welcome and Introductions 9:00-11:00

  • Location: Siemens AI Lab Munich
  • 9:00-9:30 Arrivals and Breakfast
  • 9:30-10:00 Florian Buettner - Siemens AI Residency
  • 10:00-10:45 Markus Geipel - Challenges with Industrial Machine Learning

The role of Uncertainties in Applied Machine Learning 11:00-12:30

  • Location: Conference Room Hotel Louis
  • 11:00-11:45 Neill Campbell - University of Bath
  • 11:45-12:30 Florian Buettner - Siemens

Lunch

There will be no organised lunch, but there are plenty of nice places just outside from the workshop location.

Models of Composite Uncertainty 14:00-18:00

  • Location: Conference Room Hotel Louis
  • 14:00-14:45 Marc Deisenroth - University College London - The Role of Uncertainty in Model-based Reinforcement Learning
  • 14:45-15:30 Markus Kaiser - Siemens and TU Munich - Uncertainties Need a Purpose
  • 15:30-16:00 Coffee Break
  • 16:00-16:30 Alexandra Gessner - MPI Tuebingen - Efficient integration of linearly constrained Gaussians
  • 16:30-17:15 Maren Mahsereci- Amazon Research - Challenges in Bayesian Quadrature
  • 17:15-17:45 Erik Bodin - University of Bristol - Surrogate Models for Bayesian Optimisation

Workshop Dinner

Friday 11th of October

Modelling and Methodology for Uncertainty Propagation 9:00-13:00

  • Location: Conference Room Hotel Louis
  • 9:00-9:45 Andreas Damianou - Amazon Research
  • 9:45-10:30 Ieva Kazlauskaite - University of Bath - Composite Uncertainty
  • 10:30-11:00 Coffee Break
  • 11:00-11:45 Zhenwen Dai - Amazon Research
  • 11:45-12:15 Hans Kersting - MPI Tuebingen - Convergence Rates of Gaussian ODE Filters
  • 12:15-12:45 Athenais Gautier- IDIAP - Goal-Oriented adaptive Sampling under Random Field Modelling of response Distributions

Lunch

There will be no organised lunch, but there are plenty of nice places just outside from the workshop location.

Modelling and Measures of Success 14:00-16:00

  • Location: Conference Room Gotel Louis
  • 14:00-14:45 Mark van der Wilk - Prowler
  • 14:45-15:30 Steffen Udluft - Siemens
  • 15:30-16:00 Coffee Break

Conclusions and Future Directions 17:00-18:00

  • Location: Conference Room Gotel Louis

Information

Travel

From the Airport

The international airport is located about 20km from the city center but is well connected by public transport. To get to the city center, buy a ticket in the wending machines at the airport (11.90 Euros) and take either S8 or S1 to the center of town. The quickest way to get to the city center is to take S8 to Ostbahnhof and then change to the U-Bahn which can take you anywhere in the city.

From the Trainstation

Its a nice 20 minute walk from the trainstation to the workshop location or you can take any S-Bahn going towards Ostbanhof and get off at Marienplatz from where it is a short walk.

Hotels

Munich city center where the workshop is located is quite expensive, however public transport is cheap and reliable so it is easy to stay anywhere close to a U-Bahn station and travel to the workshop location.