About

Learn about the COMPASS project

COMPASS is a scientific research project that intends to develop a unified framework for signal processing and machine learning that can evaluate and improve large-scale, multi-modal time-series data in spite of low resolution, noise, and missing values. COMPASS will handle high-dimensional incomplete inputs, solve inverse problems under a few or zero-shot performance, quantify uncertainty and detect hallucinations for reliable predictions by incorporating advanced deep learning into a generative-discriminative model. Large-scale distributed learning and a real-time platform that uses low-quality data to initiate focused high-resolution monitoring will be made possible by the framework's application to Earth Observation.

Project's Objective

Objective 1

Tackle inverse issues without depending on large amount of training data, combining generative and discriminative DL models and

Objective 2

Expand high-dimensional neural network frameworks to incorporate generative elements, allowing for unified processing of multi-temporal and multi-spectral data.

Objective 3

Quantify both in-domain and out-of-domain uncertainty in inverse problems, enhancing model reliability and address anomaly and hallucination detection in high-dimensional data.

Objective 4

Modify computational sensing framework detecting anomalous environmental changes, optimizing large-scale and federated training for EO data, and providing FAIR-compliant, high-quality datasets.

Objective 5

Develop a real-time online platform for land cover and extreme event detection, integrating low- and high-resolution EO data to guide targeted observations. The system is going to visualize alerts, historical data, and risk indices, enabling dynamic monitoring and rapid disaster response.

Page updated

Google Sites

Report abuse