Year: 2018
Type: dissertation
Abstract: Over the last few years, neural networks have become extremely popular, and their usage\nis increasing rapidly. This project has investigated the use of neural networks for one-step\ntime series forecasting on highly random data. Multi-layer perceptron (MLP), convolutional\nneural networks (CNN), recurrent neural networks (RNN), and long short-term\nmemory (LSTM) cells are tested to see if they can give a binary classification accuracy\nabove 50% using this data. The assignment focuses on designing a small embedded neural\nnetwork with low latency.\n\nThe different neural network architectures are built using a deep learning library in\nPython, called Keras. This is a high-level software framework, built on top of either Tensorflow or Theano, for fast and easy prototyping of neural networks.\n\nThe conclusion of the study is that only the CNN satisfied the requirements of the\nassignment during the work of this thesis. None of the other architectures showed sign\nof learning generalized patterns and structures from the dataset in question. The CNN\nshowed the most promising results, being able to extract information about the training\nset that increased the classification accuracy of the test. This leads the way for further\ndevelopment and an eventual hardware implementation of the inference phase reducing\nthe run-time latency.
Source: BIBSYS Brage (BIBSYS (Norway))
Author: Sigurd Øyen
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2889281484