ε-MC nets: a compact representation scheme for large cooperative game settingsε-MC nets: a compact representation scheme for large cooperative game settings Πλήρης Δημοσίευση σε Συνέδριο Conference Full Paper 2024-12-042022enIn this paper we put forward ε-MC nets, a novel succinct rule-based representation scheme for large cooperative games. First, we provide a polynomial algorithm that reaches the proposed representation by exploiting the agents’ estimates over marginal contributions, along with their acceptable information loss, ε, regarding these estimates. Then we introduce the notion of equivalence classes of agents, and exploit it to (i) obtain an even more compact representation; and (ii) derive new, previously unheld, beliefs over the value of unobserved agent collaboration patterns. Moreover, we present theoretical and empirical results on the information loss arising from this “representational compression”, and on the degree of succinctness achieved. Notably, we show that an arbitrary number of merges to reach the compressed representation, exhibits an information loss that does not exceed ε. Finally, we provide theoretical guarantees for the coalitional relative error and the Shapley value in the ε-MC net with respect to the initial representation.E. Streviniotis has been supported by the Onassis Foundation - Scholarship ID: G ZR 012-1/2021-2022.http://creativecommons.org/licenses/by/4.0/178–19015th International Conference on Knowledge Science, Engineering and ManagementKnowledge Science, Engineering and Management Streviniotis Errikos Στρεβινιωτης Ερρικος Georgara Athina Γεωργαρα Αθηνα Chalkiadakis Georgios Χαλκιαδακης Γεωργιος Springer Knowledge representation Large coalitional games MC nets Rule-based representation Equivalent agents