Design and implementation of parallel SOM model on GPGPU

Parallel implementation of neural networks is amongst major area of research in computer science. Self Organizing Map (SOM) is a neural network that has been under spotlight throughout last decade for implementation in parallel architecture. SOM trains itself through unsupervised learning by retrieving inherent topological features of applied input data. In this paper design and implementation of a parallel SOM model for GPGPU is presented. This paper focuses on CPU- GPGPU combination using CUDA platform for software development of SOM algorithm. The images of different N × N dimensions are feed as input to the SOM network and image clustering is achieved through SOM training in the form of final weight matrix. The simulations are separately performed on CPU and GPGPU. The implementation of SOM model on GPGPU shows a decline in the overall complexity of SOM training algorithm from O(n4) to O(n3)/p' with respect to sequential implementation and a speedup maximum of 5.43 approx. for applied input with large data size.