©

Prof. Dr. Jörg Lücke: "Novel Learning Algorithms for Efficient Unsupervised and Semi-Supervised Networks"

01. Jun

Im Rahmen des Forschungskolloquiums Wirtschaftsinformatik und Data Science referiert Prof. Dr. Jörg Lücke, Professor für maschinelles Lernen an der Universität Oldenburg über "Novel Learning Algorithms for Efficient Unsupervised and Semi-Supervised Networks".

Datum und Ort: 1. Juni 2018 12.15 Uhr C 14.102a

 

Abstract:

Ever increasing amounts of data require increasingly efficient learning algorithms. Furthermore, it becomes increasingly difficult to provide sufficient amounts of labels for supervised learning. Unsupervised and semi-supervised algorithms have, therefore, been increasingly intensively studied in recent years. I will first discuss some general challenges for the training of semi-supervised classifiers, before I introduce a novel neural network for semi-supervised learning which is derived from maximum likelihood principles. I briefly elaborate on the distinguishing features of the network, and show results on standard benchmarks. The network's performance is then critically compared to other approaches - and I discuss the general requirement of an empirical Occam's Razor for such comparisons. Following the discussion of semi-supervised classifiers, the last part of my talk will discuss efficient and ‘black box’ unsupervised learning. Applications to basic tasks like clustering and to more complex Bayesian Networks are presented and future research directions are sketched out.

Contact

Prof. Dr. Burk­hardt Funk
Uni­ver­si­täts­al­lee 1, C4.307
21335 Lüne­burg
Fon +49.4131.677-1593
burkhardt.​funk@leuphana.​de