Site Specific CS frontpage Featured Public/media

Frederik Hvilshøj PhD defence of "Fast and Explainable Deep Neural Networks"

Algorithms, deep leaning, and xAI in the air – a deep dive into the explainability of deep learning models.

Info about event

Time

Thursday 9 December 2021,  at 13:00 - 15:00

Location

Building 5342, room Ada-333

Price

Free DKK

Title of PhD thesis: Fast and Explainable Deep Neural Networks

With the increasing complexity of neural networks, an important question arises; how can we explain how such networks draw their conclusions? In Frederik Hvilshøj’s thesis, the question is considered from multiple angles. The focus of the thesis is to combine the power of deep generative models with explainability.  It covers both algorithm papers, which demonstrate how to speed up computations of central components used in generative models, and it present papers revolving around explanation techniques that utilize counterfactual examples.

The new research findings contribute to pushing the limits of explainable AI in terms of both more efficient algorithms and new methods for generating and evaluating counterfactual examples.

The PhD study was completed at the Computer Science department (a DIGIT project), Faculty of Natural Sciences, Aarhus University.
This summary was prepared by the PhD student.

Members of the assessment committee:
Professor Ole Winther, Department of Applied Mathematics and Computer Science, DTU Compute, Denmark
Professor Tijl de Bie, Department of Electronics and Information Systems, Ghent University, Belgium
Professor Susanne Bødker (chairman), Department of Computer Science, Aarhus University, Denmark

Main supervisor:
Professor Ira Assent, Department of Computer Science, Aarhus University, Denmark
Co-supervisor:
Professor Alexandros Iosifidis, Department of Electrical and Computer Engineering, Aarhus University, Denmark

Language: The PhD dissertation will be defended in English

Place:The PhD defence will partly be held online. To receive a link to the event, please send an e-mail to Ira Assent (ira@cs.au.dk) or Frederik Hvilshøj (fhvilshoj@cs.au.dk).
The defence is public.
The PhD thesis is available for reading at the Graduate School of Natural Sciences/GSNS, Katrinebjergvej 89F, Building 5132, 8200 Aarhus N.