GPU-based parallel implementation of modular CMAC neural networks

OData support
Dr. Horváth Gábor
Department of Measurement and Information Systems

My thesis is about investigation the possibilities for implementation of special CMAC networks on a graphical card. Focusing on the implementation of kernel, SOP and hierarchical CMAC network on a graphical card.

I am going to review the synthesis of special CMAC networks. I will show their benefits and drawbacks and how to teach.

The implementation is built on the nVidia CUDA programming architecture. The CUDA programming architecture will be shown too.

The implementation of the network is built on Matlab and CUDA codes in Matlab mex files. The network was realized in Matlab and the functions wanted to be accelerated had been run on the graphical card. I will show how to embed CUDA codes into mex files and how they can be compiled. I will show the handling and the working of the implementation in Matlab too.

I will show the functions speeded up by the help of CUDA. I will show what kind of kernel functions I had implemented and their working and connection.

I compare the implementation in CUDA and the implementation in Matlab.

At last I summarize how to develop the existing solution.


Please sign in to download the files of this thesis.