Presentation
Deep learning works really well for some problems. However, performance heavily depends on a large number of poorly understood hyperparameters and architecture choices. There is no satisfactory theory to rely on when making decisions, so trial and error is often required. Improved theory could significantly reduce the amount of engineering effort required when applying deep learning to new domains.
In contrast to classical machine learning models, deep learning models are often overparameterized. This means there are many ways to fit the training data (near) perfectly. The "implicit bias" determines which of these ways is chosen by a deep learning methodology. This talk argues that understanding the implicit bias of deep learning is central to developing a satisfactory theory, and discusses recent research on this topic.
Speaker
Johan Sokrates Wind is one of the dScience PhD candidates at the University of Oslo. He is part of the Computational Mathematics group. His research is mostly concerned with trying to understand the implicit bias in deep learning.
Program
11:30 – Doors open and lunch is served
12:00 – "Implicit Bias in Deep Learning" by Johan Sokrates Wind (PhD Candidate, Department of Mathematics)
This event is open for all PhD candidates and postdocs. No registration needed.
About the seminar series
Once a month, dScience will invite you to join us for lunch, soft drinks and professional talks at the Science Library. In addition to these, we will serve lunch to PhD candidates in our lounge in Kristine Bonnevies hus every Thursday. Due to limited space (40 people), this will be first come, first served. See how to find us here (download).
Our lounge can also be booked by PhDs and Postdocs on a regular basis, whether it is for a meeting or just to hang out – we have fresh coffee all day long! Read more about the seminar series here.