BEGIN:VCALENDAR
VERSION:2.0
PRODID:https://github.com/derhansen/sf_event_mgt
METHOD:PUBLISH
BEGIN:VEVENT
UID:11359-8511@www.uni-passau.de
CLASS: PUBLIC
SUMMARY:PICAIS Fellow-Talk: Representing 4D Dynamics in Machine Learning
DESCRIPTION:The grand challenge in modelling environmental and complex phys
 ical systems lies in navigating the Information Bottleneck: how can we com
 press vast, 4D spatiotemporal dynamics into the representations suitable f
 or Machine Learning (ML) models without sacrificing critical predictive po
 wer? This seminar addresses this fundamental trade-off. We will discuss th
 e strategies for Encoding high-resolution data (e.g., using autoencoders o
 r local feature extraction) and assess methods for principled Dimensionali
 ty Reduction (such as advanced PCA or manifold learning) to combat the cur
 se of dimensionality. Crucially, the talk will focus on moving beyond blac
 k-box predictions by emphasizing model Interpretation and leveraging Physi
 cs Understanding to ensure the compressed feature space is not only effici
 ent but physically meaningful, leading to models that are both accurate an
 d trustworthy.
DTSTAMP:20260324T100302Z
DTSTART:20260325T130000Z
DTEND:20260325T140000Z
END:VEVENT
END:VCALENDAR