Author : Neda Mohammadi [ Special \\\"9thSASTech-2015, Iran\\\"-2015] [Page No : 39-43] [2015]

Nearest Neighbor Search Algorithm has many applications in various sciences, for example KNN classification techniques are used often in industry and in many scientific applications. It has been applied in areas such as medical imaging, entropy estimation, data mining, machine learning and content based image retrieval. The high computational complexity in nearest neighbor algorithm is a challenge for runtime. Although presenting and solving of this problem and for small data is easy, when the database is large The fundamental problem in fast processing of data occurs. In areas such as, data mining where the nearest neighbor search algorithm is applied for it, several technologies have been used to classify the data. Several technologies in order to data classification are introduced which with increasing amount of data choosing appropriate technology for classifying them is important. CUDA technology was provided by NVIDIA and also this technology provided an opportunity for developers so that by using of their system graphics card, data in parallel they performed minimal cost and easy computational processing data. The concept of GPGPU and CUDA technology for nearest neighbor search is used. We will compare parallelism implementation of this algorithm on GPU With accessing to it’s shared memory with serial implementation of algorithm on CPU and while the program without access to shared memory on a graphics processing unit runs. It is shown that Parallel implementation of the algorithm on GPU with accessing to the shared memory in compared to the other methods discussed here is heavy computational processes in parallel method.