In this unit, some of the most widely used regression, prediction and classification models will be covered. Neural networks will be introduced, and the backpropagation as the primary training algorithm will be demonstrated. Various forms of deep neural networks such as multilayer perceptions, convolutional neural networks, and recurrent neural networks will be examined. The mathematics of stochastic optimisation is used to explain the behaviour and training of these networks. Various programming approaches will be discussed and demonstrated for the training and deployment of neural networks. The application of deep learning technologies will be discussed in areas such as pattern recognition.

Unit details

Location:
City Campus
Footscray Park
VU Sydney
Study level:
Postgraduate
Credit points:
12
Unit code:
NIT6004

Prerequisites

NIT5150 - Advanced Object Oriented Programming

Get help

Find a different unit

Learning Outcomes

On successful completion of this unit, students will be able to:
  1. Critically review the performance and applications of neural network and deep learning techniques;  
  2. Implement a systematic approach to design and evaluate neural network architecture;  
  3. Interpret relevant mathematical equations or statistical methodologies in terms of neural network architecture and deep learning methods;  
  4. Investigate and apply knowledge discovery processes and associated models to innovate deep learning applications considering the importance of data privacy and professional ethics to support and provide business solutions; and  
  5. Extrapolate knowledge and skills to design, develop, and evaluate a variety of deep learning tasks: modelling, clustering, dimensionality reduction, regression or classification.  

Assessment

Assessment type Description Grade
Laboratory Work Lab submissions (2) 20%
Project Projects (2) 40%
Case Study In-class Problem Solving Case Study 40%
The project is due in week 8 in the delivery block. Project 1 - Implement a neural network to solve the classification problem, with given benchmark dataset, motivated by the efficient and speedy decision-making nature of the neural network (such as image classification). Project 2 - Build a deep neural network from scratch and train it on a specific benchmark dataset, and then test it for detection or prediction purposes.

Where to next?

As part of a course

This unit is studied as part of the following courses. Refer to the course page for information on how to apply for the course.

VU takes care to ensure the accuracy of this unit information, but reserves the right to change or withdraw courses offered at any time. Please check that unit information is current with the Student Contact Centre.