Master's thesis Designing Large-Scale Satellite-Agnostic Geospatial Foundation Models
Master's thesis
Ref. 2024_001
In this project, you will develop and evaluate novel ways to build satellite-agnostic geospatial foundation models that are capable of processing multi-resolution data from diverse sensors during pretraining and finetuning in a computationally efficient way. The project is executed together with the Photogrammetry and Remote Sensing Group at ETH Zurich and requires an affiliation with ETH. The project provides access to IBM HPC infrastructure.
Significant progress in the development of highly adaptable and reusable artificial intelligence (AI) models is expected to have a major impact on earth science and remote sensing. Foundation models are pre-trained on large unlabeled datasets via self-supervision and subsequently fine-tuned with small, labeled datasets for various downstream tasks. The abundance of unlabeled data from multiple satellites makes remote sensing an ideal domain for pre-training large-scale foundation models. However, recent foundation models for geospatial analysis and remote sensing applications exploit only a fraction of the available bands, typically from a single satellite or at constant resolutions. Consequently, the scientific community is increasingly interested in developing approaches to build interoperable (i.e., satellite-agnostic) models for Earth observation and remote sensing that efficiently exploit multi-sensor, multi-resolution data.
Goal
This project aims to explore transformer models to result in interoperable, multi-resolution geospatial foundation models that encode information from various bands. This is an essential step in further enhancing AI models for remote sensing to unlock diverse environmental applications.
Qualifications
- Bachelor’s degree in computer science, deep learning or a related technical field, including equivalent practical experience
- Python, PyTorch, Linux, GitHub/Lab
- Problem-solving mindset: ability to explore technical literature, cross-referencing relevant information, synthesis of novel ML architectures
- Ways to stand out: hands-on experience with HPC systems and job scheduling, prior encounters with large-scale transformer architectures, especially with positional encodings, etc.
Diversity
IBM is committed to diversity at the workplace. With us you will find an open, multicultural environment. Excellent flexible working arrangements enable all genders to strike the desired balance between their professional development and their personal lives.
How to apply
If you are interested in this exciting position, please submit your application through the link below.
Apply
Rüschlikon / CH-8803 Rüschlikon
