Partial Information

This research area investigates the philosophical and formal implications of differentiating between hard (knowledge-like, truth-preserving) and soft (belief-like, ampliative) forms of logical information. The focus lies on extending traditional conceptions of logic, generally grounded on hard information, to incorporate non-monotonic and dynamic epistemic phenomena that characterize soft information. This distinction matters for understanding and modeling diverse styles of inference, including defeasible, belief-based, and multi-agent information dynamics, thereby broadening the scope and relevance of logic in contemporary contexts.

This theme encompasses methodological advancements and theoretical frameworks addressing scenarios where complete or perfect statistical knowledge about data is unavailable. It is central for the design and evaluation of privacy-preserving data disclosure methods, measuring information leakage under adversaries lacking full distributional knowledge, and integrating partial (linear) probabilistic information in decision processes such as insurance risk assessment. Research within this theme innovates in extending classical measures of information and uncertainty to accommodate uncertainty and partiality in the underlying distributions, enabling more realistic and robust privacy and decision models.

This research direction focuses on challenges and methodologies related to modeling and inference when systems or data are only partially observable or when data comprise mixtures of qualitative and quantitative variables. It encompasses theoretical developments in estimating conditional mutual information from mixed data types, constructing models from incomplete measurements, and frameworks for aggregating approximate information from agents or sensors with limited perceptual capabilities. Such works enable more accurate modeling, reasoning, and decision-making in realistic scenarios featuring partial knowledge and heterogeneous data sources.