Parallelizing Affinity Propagation Using Graphics Processing Units for Spatial Cluster Analysis over Big Geospatial...

Area of Science:

• Geographic Information Science (GIScience)
• Machine Learning
• High-Performance Computing

Background:

• Affinity Propagation (AP) is an unsupervised machine learning algorithm for classification, introduced in 2007.
• AP has seen limited application in geospatial contexts due to computational bottlenecks with large datasets.
• Traditional serial AP algorithms are time-consuming for extensive geocomputation.

Purpose of the Study:

• To address the scalability limitations of AP for large geospatial datasets.
• To introduce a parallelized AP algorithm optimized for graphics processing units (GPUs).
• To demonstrate the potential for processing big geospatial data using GPU-accelerated AP.

Main Methods:

• Parallelization of the Affinity Propagation algorithm.
• Implementation on graphics processing units (GPUs) to leverage multicore and manycore architectures.
• Exploitation of task and data parallelism for efficient geocomputation.

Main Results:

• Successful parallelization of the AP algorithm on GPUs for spatial cluster analysis.
• Demonstrated potential for significantly faster processing of large geospatial datasets compared to serial methods.
• Validation of the approach for scalable geocomputation.

Conclusions:

• GPU-accelerated AP offers a viable solution for analyzing big geospatial data.
• The proposed method enhances the applicability of AP in GIScience.
• This work paves the way for more efficient spatial cluster analysis in large-scale geospatial applications.