Quantum Computing (QC) is rapidly emerging as an alternative to traditional classical computing thanks to its potential advantages in terms of representational power and computing time. This activity targets to study the application and explore the potential of Quantum Machine Learning (QML) on realistic Earth Observation (EO) use cases.
Several EO scenarios are investigated. A first line of work focus on the classification of EO images by constructing a multi-task hybrid classical-quantum (MTCQ) model which executes the reconstruction and classification of EO images at once via end-to-end training. Especially, different quantum circuit ansatzes are tested to understand the correlation between their expressive power and classification accuracy. A second objective considers generative models, in particular quantum generative adversarial networks (qGANs). It aims to develop qGANs able to model the distribution of EO data allowing a wide panel of applications, including generation and composition of image. Finally, a third objective consists in exploring how quantum algorithms could help in time series data analysis with a focus on learning dynamical systems from data. This line of work investigates how Quantum encoding of dynamics can be useful to detect patterns and instabilities or anomalies.
Given the very exploratory nature of the field, this activity is implemented as a joint early-career research programme between ESA and CERN as part of the CERN Doctoral Student programme.
- Su Yeon Chang, Bertrand Le Saux, Michele Grossi, Sofia Vallecorsa, Hybrid Quantum-Classical Networks for Reconstruction and Classification of Earth Observation Images, ACAT Physics research workshop, Oct. 2022
- Su-yeon Chang, S. Vallecorsa, M. Grossi, B. Le Saux, Quantum Convolutional Circuits for Earth Observation Image Classification, IGARSS 2022, July 2022
- Su-yeon Chang, S. Vallecorsa, B. Le Saux, Quantum Machine Learning for Earth Observation Images, NeurIPS 2nd WS on Quantum Tensor Networks in Machine Learning