The GANTT and PERT charts for the project can be found below.
The objectives of WP1 are: (1) to formulate an optimisation framework inspired by the principle of ‘knowing not to know’, outputting sets of solutions equally optimal from a robustness point of view, and (2) to inject epistemic uncertainty into this new framework in both the loss functions and the constraints.
The overall objective of WP2 is to formulate learning frameworks based on the concept of epistemic AI, leveraging the results of WP1 on optimisation under epistemic uncertainty. These are: (1) an epistemic learning theory, (2) epistemic supervised learning and domain adaptation, (3) a theory of epistemic unsupervised learning focussing, in particular, on deep adversarial networks for explainable AI, (4) robust foundations for reinforcement learning with epistemic versions of both model-free and model-based algorithms, (5) epistemic inverse reinforcement learning.
The objectives of WP1 are: (1) to demonstrate science-to-technology feasibility for selected use cases in the autonomous driving setting; (2) to generate proof-of-concept level demonstrators of the proposed novel AI paradigms, including code (to be shared open-source) and algorithms; (3) to continually provide feedback on the effectiveness and efficiency of the new methodologies and algorithms developed in WP2, in an Agile development philosophy.
The objectives of WP4 are: (1) to cover the legal, contractual, financial and administrative management of the project, and establish stable relations with the EC throughout the project’s duration; (2) to exploit and disseminate the results of the project, and to manage Intellectual Properties rights; (3) to build the innovation capacity of partners, explore the full scientific, social and economic potential of Epistemic AI, and (4) to define viable a roadmap for an effective exploitation.