PhD Status Talk: Multi-agent Belief Management

Aus International Center for Computational Logic
Wechseln zu:Navigation, Suche

PhD Status Talk: Multi-agent Belief Management

Vortrag von Jonas Karge
In epistemic voting, multiple agents vote over a set of alternatives, with the goal of identifying the one that corresponds to the ground truth. The Condorcet Jury Theorem (CJT) is a foundational result in this context, offering probabilistic guarantees for truth-tracking, albeit under restrictive assumptions. This talk has two main objectives. First, we generalize the CJT by relaxing its core assumptions and provide novel probabilistic bounds for successful truth identification. Second, we introduce a new aggregation method, Voting for Bins, grounded in the epistemic voting framework, designed for aggregating interval-valued probabilistic beliefs. Each of these contributions is tied to a key application: for the generalized CJT, we formalize and bound conditions under which diversity can outperform ability, offering a rigorous formulation of the Diversity-Trumps-Ability Theorem. For Voting for Bins, we address the problem of dilation in imprecise belief aggregation.