Um motor de inferência paraconsistente para suporte à investigação sobre Degenerescência Macular da Idade

BANSKY central objective is twofold: • the development of an inference engine able to extract useful knowledge and unsuspected correlations from heterogeneous, potentially contradicting data; • and its systematic application to the study of a longer than a decade data on AMD (age-related macular degeneration), a multifactorial disease of the macula recognised as the leading cause of vision loss in western countries in people aged over 55 years. Progress in understanding AMD, through non-standard ways able to incorporate potentially contradicting or weakly consistent data in long term medical diagnosis, will have, if successful, a huge practical impact in ophthalmology and medical care. Moreover, it is expected that the reasoning techniques and the inference engine to be developed in the project will be useful in other contexts, in and out the medical domain. Actually, the world of data, pervasive to all natural and artificial ecosystems, is a mined field. Not only the values and structure of data changes from one computation to another, but also the logic under which this information needs to be understood changes as well (e.g. from classical to probabilistic, from fuzzy to linear). On the other hand, informational states may exhibit potentially inconsistent (or partially consistent) data, reflecting the diversity of judgements (e.g. from different domain experts). Moreover, it may be linked by both positive transitions (witnessing e.g. the existence of a computational step, or its cost, or its probability, etc.) and negative transitions (recording whatever prevents such a step to occur, and also possibly expressed as a specific weight). Finally, the weights of such transitions are, in most of cases, non-complementary, opening an inference arena encompassing both classical, vague and even (controlled forms of) inconsistent reasoning. The novelty of the approach proposed here comes from the decision to take both vagueness and contradiction, as ubiquitous characteristics of data, as first-class citizens in logic inference. The project will push the boundaries of state-of-the-art research on paraconsistent logics, a mathematical domain still undeveloped despite an increasing, emerging interest. New results are envisaged at different levels: from more foundational work on logic semantics and proof theory, to inference computational techniques and tool design. On the other hand, the application of the envisaged inference engine to AMD, and the systematic assessment of the inferred knowledge by specialized ophthalmic teams, at AIBILI, will provide an entirely new way to conduct data analysis and inference over concrete data repositories on AMD. This is expected to address what medical research identifies as the present bottleneck created by the lack of powerful enough reasoning frameworks to process the underlying complex and potentially contradictory empirical data.