Postdoc: Weakly supervised ML-based Earth observation
Core
Develop an advanced Earth observation (EO) machine learning framework to identify, geolocalize, and estimate physical and socio-economic impacts of climate extremes using satellite image sequences and weakly supervised learning.
Role type
Postdoctoral Researcher (Weakly Supervised ML for Earth Observation)
Builds
Annotation-efficient, weakly supervised ML-based Earth observation models with uncertainty quantification for climate adaptation.
Domain
Climate science, Earth Observation, Machine Learning, Remote Sensing
Deliverable
production ML models
Required skills
Deep learning architectures, Computer vision, Spatio-temporal modeling, Weakly supervised learning, Uncertainty quantification, PyTorch or TensorFlow, Satellite data processing
Preferred skills
Multi-spectral geospatial formats, Radar (SAR) data, Parameter-Efficient Fine-Tuning (PEFT/LoRA), Out-of-distribution detection, Change-detection tasks, Peer-reviewed publications in top-tier ML conferences
Technologies
PyTorch, TensorFlow, Sentinel-1/2, Landsat/HLS, VIIRS/MODIS, SNGP
Responsibilities
Architect data-efficient change detection modules on frozen geospatial foundation models; Engineer models to learn robust representations using minimal labeled instances and sparse text reports; Integrate geometric constraints into adapter layers to flag high epistemic uncertainty.
Seniority
Postdoctoral Researcher