Implementation of Neural Network Using Programable Devices

Ovaj diplmoski rad prolazi kroz mogućnosti implementacije neuronskih mreža na programirljivom sklopovlju. Rad će se fokusirati na korištenje FPGA pločice te navesti neke od prednosti i mana takve implementacije neuronskih mreža. Kroz praktični dio ovaj rad pokazuje neke od problema s kojima se možemo susresti tijekom implementacije neuronske mreže na FPGA pločici te pokazuje kako će mogućnost implementacije neuronske mreže na FPGA pločici ovisiti ponajviše o karakteristikama mreže koju želimo implementirati te mogućnostima FPGA pločice na kojoj to želimo napraviti. Rad pokazuje izrazito veliku važnost dobre analize problema i odabira odgovarajuće FPGA pločice kako se ne bi našli u problemima zbog fizičkih ograničenja odabranog sklopovlja.This master thesis explores ability to implement neural networks on programmable devices. This thesis focuses on the use of FPGA devices and lists some of the pros and cons of such implementation for neural networks. Through the practical part this thesis shows some of the problems we can encounter while implementing neural networks for FPGA device. It shows how our ability to implement some neural network on FPGA device will depend on characteristics of the neural network as well as of the FPGA device itself. This thesis will note that the proper analysis of our problem is of critical importance when choosing an FPGA device so that we would not find ourselves in any problems caused by physical restrictions of our device