We describe a modular elicitation framework for iterative combinatorial auctions. The framework includes proxy agents, each of which can adopt an individualized bidding language to represent partial value information of a bidder. The framework leverages algorithms from query learning to elicit value information via bids as the auction progresses. The approach reduces the multi-agent elicitation problem to isolated, single-agent learning problems, with competitive equilibrium prices used to facilitate auction clearing even without complete learning.
/lp/association-for-computing-machinery/a-modular-framework-for-iterative-combinatorial-auctions-BU8Z0Gff8A