Exploring local search in winner determination problem

This special issue includes four articles that were presented in the 15th Workshop on Information Technology and Systems (WITS’05), held in December 10–11, 2005, Las Vegas, Nevada. Since its first meeting in 1991 in MIT, WITS has become a premier annual forum for discussion and interaction among scholars with research interests in the cutting edge issues of information technology and systems. WITS’05 addressed many important research issues as the corporate world evolves toward service-oriented computing that can support business on demand, enabled by the emergence of web services, agent technologies, wireless computing, e-commerce standards, P2P technologies, the business web, among other technologies. The four selected articles have been extended substantially and went through two rounds of reviews. Collectively, these four articles explore new patterns and mechanisms that extend the new boundaries of information technologies and systems. It is our pleasure to make these four fine articles available to the community through the journal of Information Technology and Management. We thank all authors and reviewers for their diligent work that made this special issue possible. Next, we give a brief overview of the four papers in terms of their main discoveries and important contributions. The first article, ‘‘Combining probability models and Web mining models: A framework for proper name transliteration’’ by Zhou, Huang, and Chen combines probability models and web mining models to develop a framework for proper name transliteration. The authors argue that web pages are created today in almost every popular language. However, language boundaries prevent information sharing and discovery across countries. Their paper studies transliteration (referred to that proper names are translated phonetically) issue on the Web. Previous transliteration models can be categorized into four approaches: a rule-based approach, a machine learning approach, a statistical approach, and a web mining approach. This paper proposes a generic proper name transliteration framework which improves the traditional statistical approach by incorporating an enhanced Hidden Markov Model and a web mining model. Evaluation of the proposed framework is very promising as it boosted the performance by 79.05%. Singh and Sen in their article ‘‘Exploring local search in winner determination problem,’’ report their research in developing a new combinatorial auction mechanism for a wide range of practical applications such as allocation of railroads, auction of adjacent pieces of real estate, auction of airport landing slots, and distributed job shop scheduling. Determining the winning bundles in combinatorial auctions is an NP-complete problem. The inefficiency problem in existing approaches to combinatorial auction poses a limitation in conducting combinatorial auctions with multiple rounds of bidding and hinders the widespread use of combinatorial auctions on the Internet. This paper proposes a local search algorithm for combinatorial auctions, which takes only fraction of the CPU time taken by T. Han College of Business, University of Nevada, NV, Las Vegas 89154, USA

Exact Algorithms for the Matrix Bid Auction

Online auctions, including online Combinatorial Auctions, are important examples of e-commerce applications. In this paper, a Combinatorial Auction Web Platform (CAWP) is introduced. The platform enables both product selling and buying capabilities that can be realized in a combinatorial way. CAWP supports a Sealed-Bid Single-Unit type of Combinatorial Auctions. Easy customization for any selected problem domain is a distinguished feature of CAWP. Platform users are not expected to have any technical knowledge about how to solve the Winner Determination Problem (WDP) known to be critical for profit maximization of the auctioneers in Combinatorial Auctions.

Auctions, and combinatorial auctions (CAs), have been successfully employed to solve coordination problems in a wide range of application domains. However, the scale of CAs that can be optimally solved is small because of the complexity of the winner determination problem (WDP), namely of finding the bids that maximise the auctioneer’s revenue. A way of approximating the solution of a WDP is to solve its linear programming relaxation. The recently proposed Alternate Direction Dual Decomposition algorithm (AD 3) has been shown to efficiently solve large-scale LP relaxations. Hence, in this paper we show how to encode the WDP so that it can be approximated by means of AD 3. Moreover, we present PAR-AD 3, the first parallel implementation of AD 3. PAR-AD 3 shows to be up to 12.4 times faster than CPLEX in a single-thread execution, and up to 23 times faster than parallel CPLEX in an 8-core architecture. Therefore PAR-AD 3 becomes the algorithm of choice to solve large-scale WDP LP relaxations for hard instances. Furthermore, PAR-AD 3 has potential when considering large-scale coordination problems that must be solved as optimisation problems.

In this paper we present a distributed system that is capable of solving the Winner Determination Problem (WDP) for combinatorial auctions by use of a genetic algorithm. Combinatorial auctions are simple, allowing agents to bid on combination of items rather than making several individual bids. These types of auctions lead to more economical efficient allocations of goods in situations where the agents' have preference over bundles of goods. However, determining the winner(s) has been shown to be a NP-complete problem. This means that there is no simple way to solving the WDP. Our approach is to use a distributed system that solves the WDP using a genetic algorithm that is being distributed. In this way we are able to take advantage of the benefits that distributed systems offer. The results obtained from the system are compared to IBM's algorithm, known as CPLEX.

The combinatorial auction (CA) is an efficient mechanism for resource allocation in different fields, including cloud computing. It can obtain high economic efficiency and user flexibility by allowing bidders to submit bids for combinations of different items instead of only for individual items. However, the problem of allocating items among the bidders to maximize the auctioneers’ revenue, i.e., the winner determination problem (WDP), is NP-complete to solve and inapproximable. Existing works for WDPs are generally based on mathematical optimization techniques and most of them focus on the single-unit WDP, where each item only has one unit. On the contrary, few works consider the multi-unit WDP in which each item may have multiple units. Given that the multi-unit WDP is more complicated but prevalent in cloud computing, we propose leveraging machine learning (ML) techniques to develop a novel low-complexity algorithm for solving this problem with negligible revenue loss. Specifically, we model the multi-unit WDP as an augmented bipartite bid-item graph and use a graph neural network (GNN) with half-convolution operations to learn the probability of each bid belonging to the optimal allocation. To improve the sample generation efficiency and decrease the number of needed labeled instances, we propose two different sample generation processes. We also develop two novel graph-based post-processing algorithms to transform the outputs of the GNN into feasible solutions. Through simulations on both synthetic instances and a specific virtual machine (VM) allocation problem in a cloud computing platform, we validate that our proposed method can approach optimal performance with low complexity and has good generalization ability in terms of problem size and user-type distribution.

Combinatorial auction, where bidders can bid on bundles of items, has been the subject of increasing interest in recent years. Although much research work has been conducted on combinatorial auctions, most has been focusing on the winner determination problem. A largely unexplored area of research in combinatorial auctions is the design of bidding strategies, in particular, those that can be used in open and dynamic markets in which market situation is generally constantly changing. Obviously, a good bidding strategy to be used in such realistic markets cannot be obtained by analytical methods, which require all market information be known before a solution can possibly be found. Machine learning based approaches are not completely appropriate, either, as an optimal strategy learned from one market generally no longer perform well when the market situation changes. In this paper, we propose a new adaptive bidding strategy for combinatorial auction-based resource allocation problem in such dynamic markets. A bidder adopting this strategy constantly perceives the market situation, and adaptively reviews and adjusts his bid determination, thus responding to the dynamic market in a timely manner. Experiment results show that agents adopting this adaptive bidding strategy perform very well, even without any prior knowledge about the market.

Deregulation of telecommunications has meant an increase in third-party service provision, personalized service delivery and integrated networks and media. The efficient allocation of services, without human intervention, to satisfy advanced service requirements spanning several networks is a crucial task. This can be modeled as a winner determination problem in combinatorial auctions where there are multiple services, service providers and winner determination criteria (like cost, bandwidth, delay, etc) but we have shown the problem is NP-complete. This paper describes a new two-stage algorithm for optimal anytime winner determination. In the first stage, a hierarchical task network planner is used to decompose a task into subtasks that can be solved by the available services. In the second stage, a genetic algorithm with heuristics is used to find the optimal combination of service providers to provide the services identified. We show through various experiments that the genetic algorithm finds optimal solutions quicker than a modified depth-first search algorithm.

In combinatorial auctions, prospective buyers bid on bundles of items for sale, including but not limited to singleton bundles. The bid price given by a buyer on a particular bundle reflects his/her perceived utility of the bundle of items as a whole. After collecting all the bids, the auctioneer determines the revenue-maximizing assignment of winning bidders to bundles subject to nonoverlapping of bundles. To accomplish this, the auctioneer needs to solve a winner determination problem (WDP). The exactly same way of thinking can be taken to the context of min-sum scheduling, where jobs can be viewed as bidders who bid on bundles of discrete time periods on machines. Particular problems often permit only a subset of bundles. By putting appropriate restrictions on the collection of permissible bundles, we can derive from the WDP, various integer programming (IP) formulations for nonpreemptive as well as preemptive min-sum scheduling problems. We thus obtain the well-known time-indexed IP formulation in the nonpreemptive case, and further, a new strong IP formulation in the preemptive case.

Combinatorial auctions are efficient mechanisms for allocating resource in complex marketplace. Winner determination, which is NP-complete, is the core problem in combinatorial auctions. This paper proposes an immune partheno-genetic algorithm (IPGA) for solving this problem. Firstly, a zero-one programming model is built for the winner determination problem with XOR-bids and OR-bids. Then, steps of constructing three partheno-genetic operators and an immune operator are introduced. In the immune operation, new heuristics are designed for vaccines selection and vaccination. Simulation results show that the IPGA achieves good performance in large size problems and the immune operator can improve the searching ability and increase the converging speed greatly.

Continuing advances in modern technologies have transformed the procedure of transportation procurement through auctions in supply chain management (SCM). This study examines the online combinatorial auction (CA), which serves customers placed at the nodes of a transportation network, with particular consideration given to carbon emissions. The CA mechanism allows early shipments of the carriers to improve their load consolidation and reduce their repositioning trips. Sustainability and carbon emissions are considered by prioritizing the carrier’s carbon reduction performances. Two models are examined under the carbon emission regulations (Carbon tax and Cap-and-offset) to choose the winners in the CA. Our aim is to minimize the cost of transportation procurement and reduce carbon emissions by incorporating the green reputation-based winner determination problem within the procurement model. Computational experiments reveal the positive impact of prioritization and discounted offers in reducing both transportation costs and the number of empty trips. Indeed, our results show the introduction of the discounted bids allows a reduction of about 2% in the transportation cost for the shipper and 24% of empty movements, on average, for the carriers.

In this paper, on the basis of the analysis of common market model and some economic theories in the cloud computing resource management process, we propose a cloud resource management model based on combinatorial double auction. In order to solve the winner determination problem (WDP) in the combinatorial double auction, a cloud resource combinatorial double auction algorithm based on genetic algorithm and simulated annealing algorithm is proposed. Simulation results reveal that the algorithm combines genetic algorithm with simulated annealing algorithm (SAGA) outperforms genetic algorithm on fitness value and stability, and as the number of bidders increase, the solution have higher fitness value can be obtained.

An important implementation issue in the design of hybrid evolutionary multiobjective optimization algorithms with local search (i.e., multiobjective memetic algorithms) is how to strike a balance between local search and global search. If local search is applied to all individuals at every generation, almost all computation time is spent by local search. As a result, global search ability of memetic algorithms is not well utilized. We can use three ideas for decreasing the computation load of local search. One idea is to apply local search to only a small number of individuals. This idea can be implemented by introducing a local search probability, which is used to choose only a small number of initial solutions for local search from the current population. Another idea is a periodical (i.e., intermittent) use of local search. This idea can be implemented by introducing a local search interval (e.g., every 10 generations), which is used to specify when local search is applied. The other idea is an early termination of local search. Local search for each initial solution is terminated after a small number of neighbors are examined. This idea can be implemented by introducing a local search length, which is the number of examined neighbors in a series of iterated local search from a single initial solution. In this paper, we discuss the use of these three ideas to strike a local-global search balance. Through computational experiments on a two-objective 500-item knapsack problem, we compare various settings of local search such as short local search from all individuals at every generation, long local search from only a few individuals at every generation, and periodical long local search from all individuals. Global search in this paper means genetic search by crossover and mutation in multiobjective memetic algorithms.

Determining winners in combinatorial auctions is an NP-complete problem. Based on the idea of searching Nash Equilibria (NE), this paper presents a local search procedure to determine winners. To improve the solution quality calculated by the local search, we propose Nash Equilibrium Search Approach (NESA) to probe various NE solutions. According to the simulation results, auctioneer¡¦s revenue in NESA is at least 1.08% and 9.10% more than that in a general GA and Casanova, respectively. A stable solution based on NESA will be obtained within 400 seconds. Moreover, simulation results show that the solution quality of NESA is near-optimal.

On May 7, 2011, over one hundred bidders took part in a combinatorial auction for housing space in a newly erected building in Amsterdam (the Netherlands). This paper describes the development of this auction. We sketch our collaboration with the housing corporation that resulted in design choices with respect to first/second price, feedback, number of rounds, and tractability of the combinatorial auction. Furthermore, the winner determination problem is complicated by various municipal and building regulations that the allocation needs to satisfy. We show how these regulations can be included in an integer program that is used to solve the winner determination problem. Finally, computational experiments illustrate the tractability of this model.