23 January 2020 from 15:15 to 16:45

Driven by Data seminar #1: Visual Analytics for Opening the Black Box of Machine Learning

After the success of last year's topical seminars on Machine Learning, Driven by Data launches a new seminar series, this time on the theme: Discovery through Visualization. Alex Telea and Michael Behrisch will start where we left off last year and connect machine learning to visualization.

The gap between how deep learning techniques actually operate and what their users understand of their operation only widens.

Visual Analytics for Opening the Black Box of Machine Learning

In the last years, machine learning, and in particular deep learning, has witnessed enormous progresses, with increasingly more powerful techniques being developed and applied in many domains. However, as such techniques become more powerful, they also become more complex, so the gap between how they actually operate and what their users understand of their operation only widens. This leads to them being often used as black boxes, with serious consequences: When they do not work optimally, one often does not know which of their many parameters to change; and even when they work well, one wonders how much their opaque behaviour can be trusted.

Alex Telea and Michael Behrisch

Interactive visualization

In this talk we and advocate the use of visual analytics for opening this black box. We present several interactive visualization approaches that explore different facets of machine learning, enabling users to understand how such algorithms operate; what causes their suboptimal behaviour; and how to improve upon this behaviour. We demonstrate our proposals with several applications of machine learning on real-world data and problems.

About the Discovery through Visualization seminar series

At the moment large and multi-scale scientific data sets are being generated from all kinds of sensors and from model simulations. The challenge is to efficiently analyse these data with the aim to answer underlying scientific questions. Visualization is one way to do this and can be done in very different ways, from simple two-dimensional plots via multi-dimensional animations on Tiled Panel Displays to virtual reality.

This series of seminars will be focused on how to optimize discovery through the use of visualization. Experts from a range of different domains will give examples and what tools and methods they use to stimulate discovery in data. The series is organized by Driven by Data, a cluster of five UU focus areas: Bioinformatics, Foundations of Complex Systems, Applied Data Science, Game Research, and Human-centered AI.

Save the dates – speakers will be announced soon!

  • Thursday February 13, 15:15 – 16:45. Location: Ruppert D
  • Thursday March 5, 15:15 – 16:45. Location: Ruppert 002
  • Thursday March 19, 15:15 – 16:45. Location: Ruppert 002
  • Thursday April 2, 15:15 – 16:45. Location: Ruppert D
  • Thursday April 16, 15:15 – 16:45. Location: Ruppert 116
  • Thursday May 14, 15:15 – 16:45. Location: Ruppert 002
  • Thursday June 4, 15:15 – 16:45. Location: Ruppert 002
  • Thursday June 18, 15:15 – 16:45. Location: Ruppert 002
Start date and time
23 January 2020 15:15
End date and time
23 January 2020 16:45
Entrance fee
Free