Dataset Distillation - A Gentle Introduction
Practical information:
Want to register?
- Prerequisites: Familiarity with general Machine Learning
- Price: Free
Georganiseerd door:
Leertraject
We live in a world of big data. The amount of information that is available online is massive and exponentially increasing, and the datasets that are used in Machine Learning -- which allow for very impressive models -- follow suit. However, these large datasets offer their own unique problems, such as ever-increasing computational costs and enormous storage requirements. This talk will dive into the topic of dataset distillation, an approach to compress the information of a dataset in a more manageable size.
After a brief introduction on this topic and its comparison to dataset selection, a state-of-the-art method will be dissected and discussed. Afterwards, we will go over some additional advantages that emerge when compressing a dataset, as well as list some potential pitfalls associated with this technique.
Speakers
- Benjamin Vandersmissen - University of Antwerp, sqIRL/IDLab
- José Oramas - University of Antwerp, sqIRL/IDLab
Teachers / speakers
Benjamin Vandersmissen
PhD Student at University of Antwerp - Department of Computer Science
Conducting research into Sparse Representation Learning. More specifically, my research topics are related to model explainability, neural network pruning, and dataset distillation
José Oramas
José Oramas is an Assistant Professor at the Internet Data Lab (IDLab) a joint research lab between the University of Antwerp and IMEC. He received his Ph.D. at the Center for Processing Speech and Images (ESAT-PSI)of KU Leuven in April 2015. Earlier he received his engineering degree from Escuela Superior Politecnica del Litoral in Ecuador. During his Ph.D. he conducted research on understanding how groups of elements from the image (objects, object-parts, image regions, trajectories, etc.) interact and how the relationships between them can be exploited to improve artificial visual perception problems. This fueled his interest towards investigating exploratory/explanatory models that can identify informative intermediate representations and use them as means to justify the predictions that they make.
Research Interests: Representation Learning, Interpretability and Explainability, Multiple Instance Learning, Machine Learning, Deep Learning and Computer Vision
Related courses
Summer School on Security and Privacy in the Age of AI
Summerschool - Heverlee - KU Leuven, UGent, VUB, Imec