## Solving Combinatorial Auctions using Stochastic Local Search (2000)

### Cached

### Download Links

- [www.cs.ubc.ca]
- [www.aaai.org]
- [www.cs.ubc.ca]
- [people.cs.ubc.ca]
- DBLP

### Other Repositories/Bibliography

Venue: | In Proceedings of the Seventeenth National Conference on Artificial Intelligence |

Citations: | 95 - 1 self |

### BibTeX

@INPROCEEDINGS{Hoos00solvingcombinatorial,

author = {Holger H. Hoos and Craig Boutilier},

title = {Solving Combinatorial Auctions using Stochastic Local Search},

booktitle = {In Proceedings of the Seventeenth National Conference on Artificial Intelligence},

year = {2000},

pages = {22--29}

}

### Years of Citing Articles

### OpenURL

### Abstract

Combinatorial auctions (CAs) have emerged as an important model in economics and show promise as a useful tool for tackling resource allocation in AI. Unfortunately, winner determination for CAs is NP-hard and recent algorithms have difficulty with problems involving goods and bids beyond the hundreds. We apply a new stochastic local search algorithm, Casanova, to this problem, and demonstrate that it finds high quality (even optimal) solutions much faster than recently proposed methods (up to several orders of magnitude), particularly for large problems. We also propose a logical language for naturally expressing combinatorial bids in which a single logical bid corresponds to a large (often exponential) number of explicit bids. We show that Casanovaperforms much better than systematic methods on such problems. 1

### Citations

527 | Algorithm for optimal winner determination in combinatorial auctions
- Sandholm
(Show Context)
Citation Context ...[16], and algorithms for dealing with problems with special structure. 1 More recently, two proposals for applying AI-style search techniques have been used with some success for winner determination =-=[7, 17]-=-. In these proposals, the structure of bids is exploited to restrict the search---if the number of bids received is relatively sparse compared to the space of possible bids, these approaches perform m... |

320 | Computationally manageable combinatorial auctions
- Rothkopf, Pekeč, et al.
- 1998
(Show Context)
Citation Context ...6, 8] or in parallel [2, 15]) exposes the agent to certain risks (e.g., obtaining one item without the other). Combinatorial auctions (CAs) have been proposed as a means of dealing with such problems =-=[14, 16, 18]-=-. Instead of selling items individually, the seller allows bids on bundles of items, allowing bidders to deal with the entities of direct interest and avoid the risk of obtaining incomplete bundles. G... |

275 | Taming the computational complexity of combinatorial auctions: Optimal and approximate approaches
- Fujishima, Leyton-Brown, et al.
- 1999
(Show Context)
Citation Context ...[16], and algorithms for dealing with problems with special structure. 1 More recently, two proposals for applying AI-style search techniques have been used with some success for winner determination =-=[7, 17]-=-. In these proposals, the structure of bids is exploited to restrict the search---if the number of bids received is relatively sparse compared to the space of possible bids, these approaches perform m... |

221 |
Market-Based Control : A Paradigm for Distributed Resource Allocation
- Clearwater
- 1996
(Show Context)
Citation Context ..., as nontrivial multiagent systems become more prevalent, researchers are looking to market protocols such as auctions as the basis for the coordination of agent activities or for resource allocation =-=[5, 18]. When mul-=-tiple items need to be sold, standard "singleitem " auction protocols may be inappropriate, particularly when items exhibit complementarities. Specifically, when a bidder attaches a value to... |

171 |
A Combinatorial Auction Mechanism for Airport Time Slot Allocation
- Rassenti, Smith, et al.
- 1982
(Show Context)
Citation Context ...6, 8] or in parallel [2, 15]) exposes the agent to certain risks (e.g., obtaining one item without the other). Combinatorial auctions (CAs) have been proposed as a means of dealing with such problems =-=[14, 16, 18]-=-. Instead of selling items individually, the seller allows bids on bundles of items, allowing bidders to deal with the entities of direct interest and avoid the risk of obtaining incomplete bundles. G... |

141 | Auction protocols for decentralized scheduling
- Wellman, Walsh, et al.
(Show Context)
Citation Context ...ids: Comparison of time (in CPU seconds) required by CASS and Casanova for finding optimal solutions. sophisticated domains, for example, those involving a temporal component such as scheduling tasks =-=[18]-=-, and multiitem combinatorial auctions of the type explored in [11]. Acknowledgements Thanks are due to Kevin Leyton-Brown, Tuomas Sandholm, Yoav Shoham, and Moshe Tennenholtz for their comments and g... |

95 | Large-step markov chains for the traveling salesman problem
- Martin, Otto, et al.
- 1991
(Show Context)
Citation Context ...estigations will include the examination of better scoring functions, different problem distributions and the use of more sophisticated SLS techniques. In particular, Iterated Local Search algorithms =-=[12]-=- appears to hold significant promise. We also intend to explore techniques for solving problems involving schematic bids without explicit conversion to explicit form. To do this we will exploit the st... |

90 | On the run-time behaviour of stochastic local search algorithms for SAT
- Hoos
- 1999
(Show Context)
Citation Context ...borhood relation we use is analogous to that used for set packing in [4]. 3.1 Casanova Casanova is a CASLS algorithm that bears a strong resemblance to the Novelty + algorithm for SAT defined by Hoos =-=[10], one of t-=-he best-performing algorithms for solving hard SAT problems known to date (see also the Novelty algorithm of [13]). It is based on scoring each search state using the "revenue per good" of t... |

88 | Sequential auctions for the allocation of resources with complementarities
- Boutilier, Goldszmidt, et al.
- 1999
(Show Context)
Citation Context ...rport at a specific time---such that obtaining one slot is useless without the other---attaching independent values to each is difficult. Furthermore, bidding for them individually (e.g., in sequence =-=[3, 6, 8]-=- or in parallel [2, 15]) exposes the agent to certain risks (e.g., obtaining one item without the other). Combinatorial auctions (CAs) have been proposed as a means of dealing with such problems [14, ... |

60 | An algorithm for multi-unit combinatorial auctions
- Leyton-Brown, Shoham, et al.
- 2000
(Show Context)
Citation Context ...ova for finding optimal solutions. sophisticated domains, for example, those involving a temporal component such as scheduling tasks [18], and multiitem combinatorial auctions of the type explored in =-=[11]-=-. Acknowledgements Thanks are due to Kevin Leyton-Brown, Tuomas Sandholm, Yoav Shoham, and Moshe Tennenholtz for their comments and general discussion of these issues. We are especially indebted to Ke... |

52 | Evaluating las vegas algorithms: Pitfalls and remedies
- Hoos, Stützle
- 1998
(Show Context)
Citation Context ...y larger size than existing systematic methods. The nature of SLS does not permit one to offer solution quality or performance guarantees.£ Instead we adopt the empirical methodology proposed by Hoos =-=[9]-=- to evaluate the success of SLS. We also consider the use of logical languages to specify schematic bids. The CA problem is traditionally formulated by supposing that each bid is a bundle of items tog... |

28 |
Greedy local improvement and weighted set packing approximation
- Chandra, Halldorsson
- 1999
(Show Context)
Citation Context ...is on developing search strategies---or, more accurately, local improvement strategies for suboptimal solutions---that have provable quality guarantees rather than good practical applicability. 3 See =-=[1, 4]-=- for examples of such results. More practical stochastic search techniques such as tabu search and simulated annealing have been applied to related problems, but apparently not directly to weighted se... |

26 | Multi-object auctions: Sequential vs. simultaneous sales - Hausch - 1986 |

23 |
Bidding in simultaneous auctions with a constraint on exposure
- Rothkopf
- 1977
(Show Context)
Citation Context ...--such that obtaining one slot is useless without the other---attaching independent values to each is difficult. Furthermore, bidding for them individually (e.g., in sequence [3, 6, 8] or in parallel =-=[2, 15]-=-) exposes the agent to certain risks (e.g., obtaining one item without the other). Combinatorial auctions (CAs) have been proposed as a means of dealing with such problems [14, 16, 18]. Instead of sel... |

5 | On the Empirical Evaluation of Las Vegas Algorithms
- Hoos, Stützle
- 1998
(Show Context)
Citation Context ... larger size than existing systematic methods. The nature of SLS does not permit one to offer solution quality or performance guarantees. 2 Instead we adopt the empirical methodology proposed by Hoos =-=[9]-=- to evaluate the success of SLS. We also consider the use of logical languages to specify schematic bids. The CA problem is traditionally formulated by supposing that each bid is a bundle of items tog... |

5 |
Competitive equilibria in and exchange economy with indivisibilities
- Bikhchandani, Mamer
- 1997
(Show Context)
Citation Context ...ime—such that obtaining one slot is useless without the other—attaching independent values to each is difficult. Furthermore, bidding for them individually (e.g., in sequence [3, 6, 8] or in parallel =-=[2, 15]-=-) exposes the agent to certain risks (e.g., obtaining one item without the other). Combinatorial auctions (CAs) have been proposed as a means of dealing with such problems [14, 16, 18]. Instead of sel... |

4 |
A sequential auction involving assymetrically informed bidders
- Engelbrecht-Wiggans, Weber
- 1983
(Show Context)
Citation Context ...rport at a specific time---such that obtaining one slot is useless without the other---attaching independent values to each is difficult. Furthermore, bidding for them individually (e.g., in sequence =-=[3, 6, 8]-=- or in parallel [2, 15]) exposes the agent to certain risks (e.g., obtaining one item without the other). Combinatorial auctions (CAs) have been proposed as a means of dealing with such problems [14, ... |

3 |
On local search for weighted k-set packing. ESA-97
- Arkin, Hassin
- 1997
(Show Context)
Citation Context ...is on developing search strategies---or, more accurately, local improvement strategies for suboptimal solutions---that have provable quality guarantees rather than good practical applicability. 3 See =-=[1, 4]-=- for examples of such results. More practical stochastic search techniques such as tabu search and simulated annealing have been applied to related problems, but apparently not directly to weighted se... |

3 |
Evidence for invariants in local search. AAAI-97
- McAllester, Kautz, et al.
- 1997
(Show Context)
Citation Context ...bears a strong resemblance to the Novelty� algorithm for SAT defined by Hoos [10], one of the best-performing algorithms for solving hard SAT problems known to date (see also the Novelty algorithm of =-=[13]-=-). It is based on scoring each search state using the “revenue per good” of the corresponding allocation. Since each neighbor can be reached by adding a bid (and adjusting), we write sc � ��� to denot... |

1 |
Competitive equilibria in and exchangeeconomy with indivisibilities
- Bikhchandani, Mamer
- 1997
(Show Context)
Citation Context ...--such that obtaining one slot is useless without the other---attaching independent values to each is difficult. Furthermore, bidding for them individually (e.g., in sequence [3, 6, 8] or in parallel =-=[2, 15]-=-) exposes the agent to certain risks (e.g., obtaining one item without the other). Combinatorial auctions (CAs) have been proposed as a means of dealing with such problems [14, 16, 18]. Instead of sel... |

1 |
Evidence for invariants in local search. AAAI-97, 321--326
- McAllester, Kautz, et al.
- 1997
(Show Context)
Citation Context ...ears a strong resemblance to the Novelty + algorithm for SAT defined by Hoos [10], one of the best-performing algorithms for solving hard SAT problems known to date (see also the Novelty algorithm of =-=[13]). It is b-=-ased on scoring each search state using the "revenue per good" of the corresponding allocation. Since each neighbor can be reached by adding a bid (and adjusting), we write sc(b) to denote t... |