GPU-accelerated Lagrangian heuristic for multidimensional assignment problems with decomposable costs

Abstract In this paper, we describe a GPU-accelerated parallel algorithm for the axial Multidimensional Assignment Problem with Decomposable Costs (MDADC), which is one of the most fundamental formulations for data association. MDADC is known to be NP-hard and is large-dimensioned in most realistic cases; hence, heuristic solutions with qualified optimality gaps is the best one can hope for, given the state-of-knowledge. The main contribution of this paper is an efficient parallelization of the Lagrangian subgradient search algorithm specifically targeted towards the Graphics Processing Units (GPUs) based on the NVIDIA Compute Unified Device Architecture (CUDA). A GPU-accelerated Linear Assignment Problem (LAP) solver is leveraged in concert with the Lagrangian scheme for further speed-up. We also implemented a multi-GPU variant of this algorithm which maintains a good speedup profile, when tested on problems with 31 billion variables, on up to 128 GPUs.

[1]  Seif Haridi,et al.  A GPU-enabled solver for time-constrained linear sum assignment problems , 2010, 2010 The 7th International Conference on Informatics and Systems (INFOS).

[2]  Breck Baldwin,et al.  Entity-Based Cross-Document Coreferencing Using the Vector Space Model , 1998, COLING.

[3]  Tomomi Matsui,et al.  An approximation algorithm for multidimensional assignment problems minimizing the sum of squared errors , 2009, Discret. Appl. Math..

[4]  Stuart C. Shapiro,et al.  Systemic test and evaluation of a hard+soft information fusion framework: Challenges and current approaches , 2014, 17th International Conference on Information Fusion (FUSION).

[5]  Aubrey B. Poore,et al.  Multidimensional Assignment Problems Arising in Multitarget and Multisensor Tracking , 2000 .

[6]  Rakesh Nagi,et al.  The graph association problem: Mathematical models and a lagrangian heuristic , 2013 .

[7]  Panos M. Pardalos,et al.  Integer programming models for the multidimensional assignment problem with star costs , 2014, Eur. J. Oper. Res..

[8]  Frits C. R. Spieksma,et al.  Approximation Algorithms for Multi-Dimensional Assignment Problems with Decomposable Costs , 1994, Discret. Appl. Math..

[9]  Aubrey B. Poore,et al.  Multidimensional assignment formulation of data association problems arising from multitarget and multisensor tracking , 1994, Comput. Optim. Appl..

[10]  Frits C. R. Spieksma,et al.  Local search heuristics for multi-index assignment problems with decomposable costs , 2004, J. Oper. Res. Soc..

[11]  Rakesh Nagi,et al.  GPU-accelerated Hungarian algorithms for the Linear Assignment Problem , 2016, Parallel Comput..

[12]  Aubrey B. Poore,et al.  A New Lagrangian Relaxation Based Algorithm for a Class of Multidimensional Assignment Problems , 1997, Comput. Optim. Appl..

[13]  Aubrey B. Poore,et al.  A Lagrangian Relaxation Algorithm for Multidimensional Assignment Problems Arising from Multitarget Tracking , 1993, SIAM J. Optim..

[14]  Sushant S. Khopkar,et al.  Data association and graph analytical processing of hard and soft intelligence data , 2013, Proceedings of the 16th International Conference on Information Fusion.

[15]  Panos M. Pardalos,et al.  Graph partitions for the multidimensional assignment problem , 2013, Computational Optimization and Applications.

[16]  Rakesh Nagi,et al.  Test and evaluation of data association algorithms in hard+soft data fusion , 2014, 17th International Conference on Information Fusion (FUSION).

[17]  Rakesh Nagi,et al.  A map-reduce lagrangian heuristic for multidimensional assignment problems with decomposable costs , 2013, Parallel Comput..

[18]  Alexander J. Robertson,et al.  A Set of Greedy Randomized Adaptive Local Search Procedure (GRASP) Implementations for the Multidimensional Assignment Problem , 2001, Comput. Optim. Appl..

[19]  Bodo Rosenhahn,et al.  Bipartite Graph Matching Computation on GPU , 2009, EMMCVPR.