Figure 1
From: Designing all-pay auctions using deep learning and multi-agent simulation

Diagram of the contest design process (Algorithm 2). Step (1) Simulate contests and agent learning to determine the utility of possible designs. Each data point represents a (contest design, auctioneer utility) pair. Step (2) Fit a deep network to predict auctioneer utility given contest design. Step (3) Optimize the output (utility) over the input (contest design) of the deep network to find the optimal design.