Online learning in online auctions

We consider here the problem of revenue maximization in online auctions, that is, auctions in which bids are received and dealt with one-by-one. In this note, we demonstrate that results from online learning can be usefully applied in this context, and we derive a new auction which substantially improves upon the performance of previous auctions for this problem. We are primarily concerned with auctions, for a single good available in unlimited supply, often described as a digital good, though our techniques may also be useful for the case of limited supply. The problem of designing online auctions for digital goods was first described by Bar- Yossef et al. [3], one of a number of recent papers interested in analyzing revenue-maximizing auctions without making statistical assumptions about the bidders who participate in the auction [5, 6, 4, 2].