## CABOB: A Fast Optimal Algorithm for Winner Determination in Combinatorial Auctions (2005)

Citations: | 49 - 9 self |

### BibTeX

@MISC{Sandholm05cabob:a,

author = {Tuomas Sandholm and Subhash Suri and Andrew Gilpin and David Levine},

title = { CABOB: A Fast Optimal Algorithm for Winner Determination in Combinatorial Auctions},

year = {2005}

}

### Years of Citing Articles

### OpenURL

### Abstract

Combinatorial auctions where bidders can bid on bundles of items can lead to more economically efficient allocations, but determining the winners is NP-complete and inapproximable. We present CABOB, a sophisticated optimal search algorithm for the problem. It uses decomposition techniques, upper and lower bounding (also across components), elaborate and dynamically chosen bid-ordering heuristics, and a host of structural observations. CABOB attempts to capture structure in any instance without making assumptions about the instance distribution. Experiments against the fastest prior algorithm, CPLEX 8.0, show that CABOB is often faster, seldom drastically slower, and in many cases drastically faster—especially in cases with structure. CABOB’s search runs in linear space and has significantly better anytime performance than CPLEX. We also uncover interesting aspects of the problem itself. First, problems with short bids, which were hard for the first generation of specialized algorithms, are easy. Second, almost all of the CATS distributions are easy, and the run time is virtually unaffected by the number of goods. Third, we test several random restart strategies, showing that they do not help on this problem—the run-time distribution does not have a heavy tail.

### Citations

3711 |
Artificial Intelligence: A Modern Approach (Second Edition
- Russell, Norvig
- 2003
(Show Context)
Citation Context ...d the optimal solution. In our experiments we only show cases where CPLEX was able to run in RAM. Everything else being equal, DFBnB 14 Best-bound search is identical to A* search (Hart et al., 1968; =-=Russell and Norvig, 1995-=-) if the next node to expand is always chosen to be the node with the greatest f = g + h. However, to avoid solving a node’s LP when the node is first seen (and only solving the LP when the node comes... |

1433 |
Reducibility among combinatorial problems
- Karp
- 1972
(Show Context)
Citation Context ...evenue under the constraint that each item can be allocated to at most one bidder: 1 max n� j=1 pjxj s.t. � j|i∈Sj xj ≤ 1, ∀i ∈ {1..m} xj ∈ {0, 1} This problem is N P-complete (Rothkopf et al., 1998; =-=Karp, 1972-=-). The problem cannot even be approximated to a ratio of n 1−ɛ in polynomial time, for any fixed ɛ > 0 (unless ZPP = N P) (Sandholm, 2002a). 2 If bids could be accepted partially, the problem would be... |

980 |
A formal basis for the heuristic determination of minimum cost paths
- Hart, Nilsson, et al.
- 1968
(Show Context)
Citation Context ...pped before it found the optimal solution. In our experiments we only show cases where CPLEX was able to run in RAM. Everything else being equal, DFBnB 14 Best-bound search is identical to A* search (=-=Hart et al., 1968-=-; Russell and Norvig, 1995) if the next node to expand is always chosen to be the node with the greatest f = g + h. However, to avoid solving a node’s LP when the node is first seen (and only solving ... |

314 | Boosting combinatorial search through randomization - Gomes, Selman, et al. - 1998 |

313 | Computationally manageable combinatorial auctions
- Rothkopf, Pekeč, et al.
- 1998
(Show Context)
Citation Context ...000; Zurel and Nisan, 2001; Anandalingam et al., 2002), and 3) restricting the bundles on which bids can be submitted so severely that the remaining problem can be solved optimally and provably fast (=-=Rothkopf et al., 1998-=-; Sandholm and Suri, 2003; Tennenholtz, 2000; Penn and Tennenholtz, 2000; van Hoesel and Müller, 2001; Lehmann et al.). The third approach suffers from similar economic inefficiencies and exposure pro... |

281 |
Integer Programming
- Wolsey
- 1998
(Show Context)
Citation Context ...st the state-of-the-art general-purpose solver. There are some fundamental differences between CABOB and CPLEX that we want to explain to put the experiments in context. CPLEX uses best-bound search (=-=Wolsey, 1998-=-), 14,15 which requires exponential space (CPLEX also has an option to force depth-first search, but that makes the search slower), while CABOB uses depth-first branch-and-bound (DFBnB) which runs in ... |

268 | Taming the Computational Complexity of Combinatorial Auctions: Optimal and Approximate Approaches
- Fujishima, Leyton-Brown, et al.
- 1999
(Show Context)
Citation Context ...ems (Rassenti et al., 1982; Sandholm, 1993) allow bidders to express complementarity among the items (and, with a rich enough bidding language, also substitutability among the items (Sandholm, 2002a; =-=Fujishima et al., 1999-=-; Sandholm, 2002b; Nisan, 2000; Hoos and Boutilier, 2001)). Perhaps the most well known combinatorial auction is the Federal Communications Commission’s planned combinatorial auction for spectrum lice... |

264 |
An implementation of the contract net protocol based on marginal cost calculations
- Sandholm
- 1993
(Show Context)
Citation Context ...luations—it can be more or less. This is often the case for example in electricity markets, equities trading, bandwidth auctions (McMillan, 1994; McAfee and McMillan, 1996), transportation exchanges (=-=Sandholm, 1993-=-, 2000, 1991; Net Exchange, Inc, 2001), pollution right auctions, auctions for airport landing slots (Rassenti et al., 1982), supply chains (Sandholm, 2002b; Walsh et al., 2000; Babaioff and Nisan, 20... |

244 | Bidding and Allocation in Combinatorial Auctions
- Nisan
(Show Context)
Citation Context ...3) allow bidders to express complementarity among the items (and, with a rich enough bidding language, also substitutability among the items (Sandholm, 2002a; Fujishima et al., 1999; Sandholm, 2002b; =-=Nisan, 2000-=-; Hoos and Boutilier, 2001)). Perhaps the most well known combinatorial auction is the Federal Communications Commission’s planned combinatorial auction for spectrum licenses. Less publicized, in busi... |

216 |
Selling Spectrum Rights
- McMillan
- 1994
(Show Context)
Citation Context ...ble items that are for sale is not the sum of the individual items’ valuations—it can be more or less. This is often the case for example in electricity markets, equities trading, bandwidth auctions (=-=McMillan, 1994-=-; McAfee and McMillan, 1996), transportation exchanges (Sandholm, 1993, 2000, 1991; Net Exchange, Inc, 2001), pollution right auctions, auctions for airport landing slots (Rassenti et al., 1982), supp... |

212 | Branch-and-price: Column generation for solving huge integer programs. Operations Research 46(3):316–329 Begur
- Barnhart, EL, et al.
- 1998
(Show Context)
Citation Context ...he operations research literature (e.g., Wolsey (1998, page 99)). The idea is that the LP is least sure about these bids, so it makes sense to resolve that 13 A related approach is column generation (=-=Barnhart et al., 1998-=-). It seems best-suited when the problem contains a huge number of variables (columns). (Typically these variables are implicitly defined, such as sequences of flight legs in crew scheduling, and thus... |

185 | Computationally feasible VCG mechanisms
- Nisan, Ronen
(Show Context)
Citation Context ...dholm, 2002a), no fast algorithm can guarantee that its solution is even close to optimal). Furthermore, suboptimal winner determination generally compromises the incentive properties of the auction (=-=Nisan and Ronen, 2000-=-; Sandholm, 2002b,a). Due to these reasons, we focus on the first approach—with the understanding that on some problem instances, any algorithm within this approach will take a long time. We present a... |

175 | Analyzing the Airwaves Auction
- McAfee, McMillan
- 1996
(Show Context)
Citation Context ...re for sale is not the sum of the individual items’ valuations—it can be more or less. This is often the case for example in electricity markets, equities trading, bandwidth auctions (McMillan, 1994; =-=McAfee and McMillan, 1996-=-), transportation exchanges (Sandholm, 1993, 2000, 1991; Net Exchange, Inc, 2001), pollution right auctions, auctions for airport landing slots (Rassenti et al., 1982), supply chains (Sandholm, 2002b;... |

166 | Clique is hard to approximate within n 1−ɛ - H˚astad - 1996 |

160 |
A Combinatorial Auction Mechanism for Airport Time Slot Allocation,” The Bell
- Rassenti, Smith, et al.
- 1982
(Show Context)
Citation Context ...dth auctions (McMillan, 1994; McAfee and McMillan, 1996), transportation exchanges (Sandholm, 1993, 2000, 1991; Net Exchange, Inc, 2001), pollution right auctions, auctions for airport landing slots (=-=Rassenti et al., 1982-=-), supply chains (Sandholm, 2002b; Walsh et al., 2000; Babaioff and Nisan, 2001), and auctions for carrier-of-last-resort responsibilities for universal services (Kelly and Steinberg, 2000). Combinato... |

159 | Winner Determination in Combinatorial Auction Generalizations
- Sandholm, Suri, et al.
- 2002
(Show Context)
Citation Context ... not only on designing faster algorithms for winner determination in combinatorial auctions, but also on winner determination in combinatorial reverse auctions and exchanges (Sandholm and Suri, 2003; =-=Sandholm et al., 2002-=-; Kothari et al., 2003), as well as in combinatorial markets with additional side constraints (Sandholm and Suri, 2001). Beyond winner determination, there are several other interesting research direc... |

145 |
Linear-space best-first search
- Korf
- 1993
(Show Context)
Citation Context ...t its default setting in the experiments that we present in this paper. 16 CABOB could be converted to use another search strategy. We considered SMA* (Russell, 1992) and recursive best-first search (=-=Korf, 1993-=-). However, we decided against SMA* because it would require keeping copies of the bid graph G (one for each leaf of the search tree), which would require frequent slow copying and would use a large a... |

142 | 2006): “Combinatorial Auctions with Decreasing Marginal Utilities - Lehmann, Lehmann, et al. |

138 | Towards a universal test suite for combinatorial auction algorithms - Leyton-Brown, Shoham - 2000 |

117 | Optimal speedup of Las Vegas algorithms
- Luby, Sinclair, et al.
- 1993
(Show Context)
Citation Context ... uniform distribution (7 items per bid, 50 items, 500 bids), reported by each algorithm once per second. • Constant: Restart after every δ backtracks (Gomes et al., 1998). • Luby-Sinclair-Zuckerman: (=-=Luby et al., 1993-=-) 22 Luby et al. showed that the constant scheme above is optimal if δ is tailored to the run-time distribution, which is, unfortunately, usually not known in practice. Therefore, they constructed a s... |

116 | Solving Airline Crew Scheduling Problems by Branch-and-Cut - Hoffman, Padberg - 1993 |

115 | The communication requirements of efficient allocations and supporting Lindahl prices”, Journal of Economic Theory
- Nisan, Segal
- 2005
(Show Context)
Citation Context ...h et al., 2002; Zinkevich et al., 2003; Blum et al., 2004; Santi et al., 2004; Lahaie and Parkes, 2004). While the communication complexity of combinatorial auctions is exponential in the worst case (=-=Nisan and Segal, 2003-=-), in practice only a vanishingly small fraction of the bidders’ private valuation information needs to be elicited in order to determine the optimal allocation (and incentive compatible payments to b... |

110 | i-bundle: An efficient ascending price bundle auction
- Parkes
- 1999
(Show Context)
Citation Context ...l other interesting research directions within combinatorial auctions. Designing mechanisms (rules) for iterative combinatorial auctions is one such direction (Sandholm, 1993; DeMartini et al., 1999; =-=Parkes, 1999-=-; Parkes 25sand Ungar, 2000a,b; Wurman and Wellman, 2000; Bikhchandani et al., 2001; Ausubel and Milgrom, 2002). A more recent, more general direction is to supplement the auctioneer with an elicitor ... |

110 | Walrasian Equilibrium with Gross Substitutes - Gul, Stacchetti - 1999 |

107 | eMediator: A next generation electronic commerce server - Sandholm |

100 | Preference elicitation in combinatorial auctions
- Sandholm, Boutilier
- 2006
(Show Context)
Citation Context ...ctioneer with an elicitor that selectively elicits bids from the bidders in order to determine a good allocation of items to bidders without requiring the bidders to bid on all combinations of items (=-=Conen and Sandholm, 2001-=-, 2002b,a; Smith et al., 2002; Zinkevich et al., 2003; Blum et al., 2004; Santi et al., 2004; Lahaie and Parkes, 2004). While the communication complexity of combinatorial auctions is exponential in t... |

97 | A new and improved design for multiobject iterative auctions. Management Sci
- Kwasnica, Ledyard, et al.
- 2005
(Show Context)
Citation Context ...nation, there are several other interesting research directions within combinatorial auctions. Designing mechanisms (rules) for iterative combinatorial auctions is one such direction (Sandholm, 1993; =-=DeMartini et al., 1999-=-; Parkes, 1999; Parkes 25sand Ungar, 2000a,b; Wurman and Wellman, 2000; Bikhchandani et al., 2001; Ausubel and Milgrom, 2002). A more recent, more general direction is to supplement the auctioneer wit... |

91 | Bidding Languages for Combinatorial Auctions
- Hoos, Boutilier
(Show Context)
Citation Context ...ers to express complementarity among the items (and, with a rich enough bidding language, also substitutability among the items (Sandholm, 2002a; Fujishima et al., 1999; Sandholm, 2002b; Nisan, 2000; =-=Hoos and Boutilier, 2001-=-)). Perhaps the most well known combinatorial auction is the Federal Communications Commission’s planned combinatorial auction for spectrum licenses. Less publicized, in business-to-business commerce ... |

87 | The package assignment model - Bikhchandani, Ostroy - 2006 |

69 | Optimal solutions for multi-unit combinatorial auctions: Branch and bound heuristics
- Gonen, Lehmann
- 2000
(Show Context)
Citation Context ...thms that provably find an optimal solution but are slow on some problem instances (Sandholm, 2002a; Fujishima et al., 1999; Sandholm and Suri, 2003; Andersson et al., 2000; de Vries and Vohra, 2003; =-=Gonen and Lehmann, 2000-=-; Lehmann and Gonen, 2001; Leyton-Brown et al., 2000b; van Hoesel and Müller, 2001; Balas and Yu, 1986; Babel and Tinhofer, 1990; Babel, 1991; Balas and Xue, 1991; Nemhauser and Sigismondi, 1992; Mann... |

69 | Akba: a progressive, anonymous-price combinatorial auction
- Wurman, Wellman
(Show Context)
Citation Context ...in combinatorial auctions. Designing mechanisms (rules) for iterative combinatorial auctions is one such direction (Sandholm, 1993; DeMartini et al., 1999; Parkes, 1999; Parkes 25sand Ungar, 2000a,b; =-=Wurman and Wellman, 2000-=-; Bikhchandani et al., 2001; Ausubel and Milgrom, 2002). A more recent, more general direction is to supplement the auctioneer with an elicitor that selectively elicits bids from the bidders in order ... |

66 | Liadan Ita Oćallaghan, and Yoav Shoham. Truth revelation in approximately efficient combinatorial auctions - Lehmann |

64 | Some tractable combinatorial auctions
- Tennenholtz
- 2000
(Show Context)
Citation Context ... 2002), and 3) restricting the bundles on which bids can be submitted so severely that the remaining problem can be solved optimally and provably fast (Rothkopf et al., 1998; Sandholm and Suri, 2003; =-=Tennenholtz, 2000-=-; Penn and Tennenholtz, 2000; van Hoesel and Müller, 2001; Lehmann et al.). The third approach suffers from similar economic inefficiencies and exposure problems as noncombinatorial auctions because t... |

60 | A Combinatorial Auction with Multiple Winners for Universal Services
- Kelly, Steinberg
- 2000
(Show Context)
Citation Context ... landing slots (Rassenti et al., 1982), supply chains (Sandholm, 2002b; Walsh et al., 2000; Babaioff and Nisan, 2001), and auctions for carrier-of-last-resort responsibilities for universal services (=-=Kelly and Steinberg, 2000-=-). Combinatorial auctions where bidders can bid on bundles of items (Rassenti et al., 1982; Sandholm, 1993) allow bidders to express complementarity among the items (and, with a rich enough bidding la... |

60 | An algorithm for multi-unit combinatorial auctions
- Leyton-Brown, Shoham, et al.
- 2000
(Show Context)
Citation Context ...e slow on some problem instances (Sandholm, 2002a; Fujishima et al., 1999; Sandholm and Suri, 2003; Andersson et al., 2000; de Vries and Vohra, 2003; Gonen and Lehmann, 2000; Lehmann and Gonen, 2001; =-=Leyton-Brown et al., 2000-=-b; van Hoesel and Müller, 2001; Balas and Yu, 1986; Babel and Tinhofer, 1990; Babel, 1991; Balas and Xue, 1991; Nemhauser and Sigismondi, 1992; Mannino and Sassano, 1994; Balas and Xue, 1996; Pardalos... |

59 | Preventing strategic manipulation in iterative auctions: Proxy agents and price-adjustment - Parkes, Ungar - 2000 |

57 | Issues in Computational Vickrey Auctions - Sandholm |

57 | Finding a maximum clique in an arbitrary graph - Balas, Yu - 1986 |

54 | Efficient memory-bounded search methods
- Russell
- 1992
(Show Context)
Citation Context ...ve. Therefore, we left the backtrack parameter at its default setting in the experiments that we present in this paper. 16 CABOB could be converted to use another search strategy. We considered SMA* (=-=Russell, 1992-=-) and recursive best-first search (Korf, 1993). However, we decided against SMA* because it would require keeping copies of the bid graph G (one for each leaf of the search tree), which would require ... |

54 | Applying learning algorithms to preference elicitation - Lahaie, Parkes - 2004 |

52 | Costly valuation computation in auctions - Larson, Sandholm - 2001 |

50 | Fredrik Ygge. Integer programming for combinatorial auction winner determination - Andersson, Tenhunen - 2000 |

50 | de Vries and Rakesh V. Vohra. Combinatorial Auctions: A survey - Sven |

49 | Linear programming and Vickrey auctions - Bikhchandani, Vries, et al. |

48 | Partial-revelation VCG mechanism for combinatorial auctions - Conen, Sandholm - 2002 |

44 | The exponentiated subgradient algorithm for heuristic boolean programming
- Schuurmans, Southey, et al.
- 2001
(Show Context)
Citation Context ...hat in combinatorial auction winner determination, CPLEX performs favorably even against incomplete search algorithms—such as stochastic local search—which do not generally find the optimal solution (=-=Schuurmans et al., 2001-=-). So, when we compare CABOB against CPLEX 8.0, to our knowledge, we are comparing it against the state-of-the-art general-purpose solver. There are some fundamental differences between CABOB and CPLE... |

41 | Side constraints and non-price attributes in markets - Sandholm, Suri - 2001 |

37 | Preference elicitation and query learning
- Blum, Jackson, et al.
(Show Context)
Citation Context ...er to determine a good allocation of items to bidders without requiring the bidders to bid on all combinations of items (Conen and Sandholm, 2001, 2002b,a; Smith et al., 2002; Zinkevich et al., 2003; =-=Blum et al., 2004-=-; Santi et al., 2004; Lahaie and Parkes, 2004). While the communication complexity of combinatorial auctions is exponential in the worst case (Nisan and Segal, 2003), in practice only a vanishingly sm... |

31 | Concurrent auctions across the supply chain
- Babaioff, Nisan
- 2001
(Show Context)
Citation Context ...hanges (Sandholm, 1993, 2000, 1991; Net Exchange, Inc, 2001), pollution right auctions, auctions for airport landing slots (Rassenti et al., 1982), supply chains (Sandholm, 2002b; Walsh et al., 2000; =-=Babaioff and Nisan, 2001-=-), and auctions for carrier-of-last-resort responsibilities for universal services (Kelly and Steinberg, 2000). Combinatorial auctions where bidders can bid on bundles of items (Rassenti et al., 1982;... |

31 | On polynomial-time preference elicitation with value queries
- Blum, Zinkevich, et al.
- 2003
(Show Context)
Citation Context ... from the bidders in order to determine a good allocation of items to bidders without requiring the bidders to bid on all combinations of items (Conen and Sandholm, 2001, 2002b,a; Smith et al., 2002; =-=Zinkevich et al., 2003-=-; Blum et al., 2004; Santi et al., 2004; Lahaie and Parkes, 2004). While the communication complexity of combinatorial auctions is exponential in the worst case (Nisan and Segal, 2003), in practice on... |