Alexander Munteanu – Dimensionality and data reduction for large-scale machine learning and optimization
January 14, 2025 @ 16:00 - 17:00
CESOC continues the seminar series “My Research” this Winter term 2024/25 with a series of talks given by scientists from University of Bonn, University of Cologne and Research Centre Jülich.
This semester, we are excited to announce that the seminar series will be in a hybrid format and hosted by the different CESOC partners and it will also be streamed online (via zoom). The seminars will take place on Tuesdays on a bi-weekly basis at 16:00 hr.
on Tuesday, 14 Jan. 2025 at 16:00 (CST)
Dr. Alexander Munteanu
from the Department of Mathematics and Computer Science, Division of Informatics, University of Cologne, talking on their work
“Dimensionality and data reduction for large-scale machine learning and optimization ”
It is open to any interested person within the CESOC research disciplines (any Earth system sciences, mathematics or computer science). Please contact info[@]cesoc.net, if you would like to participate. Full Schedule is available here!
Abstract:
Continuous physical measurements at increasing speed and resolution generate enormous amounts of data. These data need to be communicated, stored and analyzed, which consumes enormous amounts of valuable and expensive resources such as telecommunication bandwidth, physical storage, computing time and energy.
At the same time, there is a lot of redundancy in the collected data, they often exhibit sparsity, inherently low intrinsic dimension and even expressive machine learning models are naturally limited in their capacity of capturing patterns from data. These properties often allow us to identify the important structures of data and to reduce the dimensionality or volume of massive data sets so as to obtain small data summaries, which we broadly call sketches.
Sketching enables the design of highly efficient data stream and distributed algorithms, while preserving the statistical accuracy of large but computationally intractable data. Sketching thus resolves the aforementioned limitations of communicating, storing and analyzing massive data, and saves valuable resources.