Digital implementation of artificial neural networks / Saffih Faycal

Saffih , Faycal (1998) Digital implementation of artificial neural networks / Saffih Faycal. Masters thesis, Universiti Malaya.

Preview

PDF (Thesis M.A.) Download (38Mb) | Preview

Abstract

This thesis is concerned with the philosophy and the strategies related to the digital implementations of artificial neural networks ANN. An open and deep insight of the biological origin of ANN as well as its tidy relation with physics through statistical mechanics and the Hopficld revolutionary concept of computational energy (See Ref (4] of Chapter 2) is given. The Very Large-Scale Integration VLSI implementation of ANN (mainly digital) with its diverse methodologies, strategies and goals also discussed. The notion of parallelism that was the comer stone of the third chapter in the processing (or the hardware) point-of-view and is enhanced in the fourth chapter to introduce the parallel learning-processing PLP notion. The Printed Circuit Board PCB technique is applied for the implementation of PLP-based neuron Bidirectional Associative Memory BAM using the cad tool of EEDIII that simulates it and shows the technique limited capabilities. The powerful Very Hardware Description Language VHDL is introduced and used to implement more suitable versions of the previous circuit and other, to be downloaded on a Field Programmable Gate Arrays FPGA chips. The PLP neuron-based BAM implementation is enhanced further by the introduction of the encodingcomparing technique as a user interface with the network as well as the expandability technique based on the suggested systolic-like architecture. Finally. the bus-based architecture technique is presented for the implementation of the Hopfield ANN.

Actions (For repository staff only : Login required)