Loading...
Please wait, while we are loading the content...
Similar Documents
Multiple-resolution divide and conquer neural networks for large-scale TSP-like energy minimization problems
| Content Provider | Semantic Scholar |
|---|---|
| Author | Noel, Steven Szu, Harold H. |
| Abstract | Here we describe a multiple-resolution divide-andconquer ANN approach for large-scale constrained optimization problems li ke the TSP. The goal of the approach is to divide a problem into sub-problems, solve them with parallel ANNs, and then combine the resulting sub-solutions. The divide-and-conquer approach is based on a mathematical principle of orthogonal division errors, which provides the criteria for optimal problem division. The resulting division minimizes the cross-correlation among sub-problems, and hence the necessary communication among them. It therefore provides a minimum-communication allocation of sub-problems to parallel processors. Moreover, the divide and conquer can be done recursively, so that it occurs at all resolutions of the data. We show how wavelets can perform the necessary multiple-resolution data clustering. The divide-and-conquer approach is particularly effective for large-scale fractal data, which exhibits clustering over a large number of resolutions. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://csis.gmu.edu/noel/pubs/1997_IJCNN.pdf |
| Alternate Webpage(s) | http://ist.gmu.edu/~csis/noel/pubs/1997_IJCNN.pdf |
| Language | English |
| Access Restriction | Open |
| Subject Keyword | Allocation Artificial neural network Central processing unit Closing (morphology) Cluster analysis Combinatorial optimization Constrained optimization Cross-correlation Diff utility Division by zero Energy minimization Exhibits as Topic Fractal Kinetics Lithium Mathematical optimization Mathematics Neural Network Simulation Parallel computing Postal Published Comment Recursion Scheduling (computing) Scheduling - HL7 Publishing Domain Wavelet statistical cluster |
| Content Type | Text |
| Resource Type | Article |
