Materials

Presentation Geoforge – Accessible location intelligence for informed decisions
by Maaz Sheikh, Ageospatial
 

Presentation Reinforcement Learning for Flood Control
by Dr. Magnus Heitzler, Heitzler Geoinformatik
 

Presentation Geometric Deep Learning & Graph Neural Networks
Collab notebook Traffic Forecasting
by Dr. Jan Svoboda, SLF Davos
 

Presentation Gaussian Splatting for 3D Reconstruction
Jupyter notebook/Python scripts Gaussian Function Properties & Fitting
by Dr. Raimund Schnürer, EPF Lausanne
 

Document Computer Vision for Damage Assessment
by Dr. Yizi Chen, ETH Zurich
 

Presentation Accompanying slides during the workshop
by Dr. Raimund Schnürer, EPF Lausanne
 

Presentations in *.pptx format contain some videos.

Abstract

Deep learning is well-suited for essential tasks in disaster management, such as modelling, optimization, simulation, navigation, and reconstruction. In this workshop, participants will gain theoretical and practical insights into various deep learning methodologies, focusing on preventing and coping with natural or man-made disasters. Introductory, the chair will provide a brief overview and showcase the latest trends in deep learning techniques for disaster management. The workshop itself is divided into two parts:

In the first part, the GeoForge platform will be presented. This platform enables users to analyse near real-time remote sensing images, supported by a large language model. Among other applications, it has been used to assess the impact of a flood event on critical infrastructure in Bangladesh.

In the second part, participants will be split into groups focusing on risk simulation, change detection, and infrastructure reconstruction. Within the groups, a deep learning methodology—such as deep reinforcement learning, graph neural networks, vision transformers, or gaussian splatting—will be illustrated using a sandbox example of a specific type of disaster. Participants will work on small tasks supported by one of the invited experts. Afterwards, participants will report their findings to the plenary session.

The workshop will conclude with a brief colloquium where future requirements will be gathered, such as sharing datasets or deploying models. Participants will have the chance to present their ideas, ask questions of experts, and share their experiences with others.

Ideally, participants have already published or plan to publish articles using deep learning methods or intend to apply deep learning methods in practice. Basic knowledge in programming and mathematics is recommended.

Event information

30 January 2025, 2:00pm – 3:30pm
Bern, Switzerland
RIMMA2025 – International Conference on Forecasting, Preparedness, Warning and Response