Task-adapted bilevel learning of flexible statistical models for imaging and vision – TASKABILE

The project TASKABILE is positioned at the interface between three different areas: inverse problems, optimisation and learning. It aims to make the framework of bilevel optimisation a paradigm for the reliable and task-dependent estimation of adaptive feature-dependent models for imaging and vision. Differently from deep-learning black boxes, the approaches described in TASKABILE are theoretically grounded, interpretable and provide a flexible tool combining statistical/variational modelling with vision-inspired optimisation in a unified way.
Its objectives are: i) the definition of task-adapted image evaluation metrics suited to assess the effectiveness of the estimated model w.r.t. the specific task considered ii) the use of the statistical framework of Generalised Gaussian Distributions (GGDs) for the flexible and adaptive description of pixel-dependent image contents iii) the development of fast, inexact and stochastic, algorithms tailored to solve the bilevel problem up to an adequte accuracy with limited computational efforts.