Human Data Science (HDS)

Colloquium

MSDSlab Meeting 17/11/2016

 Boosting: fun and delicious.

2016-11-17-14-06-40

Erik-Jan explains boosting while gesticulating wildly

2016-11-17-14-56-04

The group launches into a tense competition for the ultimate error rate

2016-11-17-14-56-16

Maarten is the cunning champion of the spiral dataset. Bon appetit!

2016-11-17-14-59-21

Emmeke easily levels her competition in the second round

This meeting was at 14:00 in room B1.09

Topic: Boosting as a way of combining weak learners into a strong learner.

Preparation: Look at this excellent visual demonstration of boosting regression trees!

Content: First, I (Erik-Jan) gave a presentation on boosting – how it emerged and how it evolved into one of the de-facto ways of ensemble learning. Then, we went to the gradient boosting playground to try to compete for lowest test error. Last, we will discuss the implications this has for our respective fields.

Bonus: I created some code for the presentation, available here (non-final)

You must be logged in to post a comment.