GP Classifier Problem Decomposition Using First-Price and Second-Price Auctions
نویسندگان
چکیده
This work details an auction-based model for problem decomposition in Genetic Programming classification. The approach builds on the population-based methodology of Genetic Programming to evolve individuals that bid high for patterns that they can correctly classify. The model returns a set of individuals that decompose the problem by way of this bidding process and is directly applicable to multi-class domains. An investigation of two auction types emphasizes the effect of auction design on the properties of the resulting solution. The work demonstrates that auctions are an effective mechanism for problem decomposition in classification problems and that Genetic Programming is an effective means of evolving the underlying bidding behaviour.
منابع مشابه
On Revenue Maximization in Second-Price Ad Auctions
Most recent papers addressing the algorithmic problem of allocating advertisement space for keywords in sponsored search auctions assume that pricing is done via a first-price auction, which does not realistically model the Generalized Second Price (GSP) auction used in practice. Towards the goal of more realistically modeling these auctions, we introduce the Second-Price Ad Auctions problem, i...
متن کاملCS364B: Frontiers in Mechanism Design Lecture #16: The Price of Anarchy in First-Price Auctions
In this lecture we pass from second-price to first-price auctions — auctions where you pay your bid. We haven’t said much about pay-as-bid auctions in this or the previous course. The reason is that we’ve been focusing on auctions that are easy to participate in and that have strong incentive guarantees, like dominant-strategy incentive-compatibility. In this part of the course, where we focus ...
متن کاملLearning Theory and Algorithms for Auctioning and Adaptation Problems
A common assumption in machine learning is that training and testing data are i.i.d. realizations from the same distribution. However, this assumption is often violated in practice: for instance test and training distributions can be related but different such as when training a face recognition system on a carefully curated data set which will be deployed in the real world. As a different exam...
متن کاملOutside options: Another reason to choose the first-price auction
In this paper we study equilibriumand experimental bidding behaviour in first-price and second price auctions with outside options. We find that bidders do respond to outside options and to variations of common knowledge about competitors’ outside options. However, overbidding in first-price auctions is significantly higher with outside options than without. First-price auctions yield more reve...
متن کاملThe Spite Motive and Equilibrium Behavior in Auctions
We study auctions where bidders have independent private values but attach a disutility to the surplus of rivals, and derive symmetric equilibria for first-price, second-price, English, and Dutch auctions. We find that equilibrium bidding is more aggressive than standard predictions. Indeed, in second-price auctions it is optimal to bid above one’s valuation; that is, bidding “frenzies” can ari...
متن کامل