Statistical Machine Learning

Subject COMP90051 (2016)

Note: This is an archived Handbook entry from 2016.

Credit Points: 12.5
Level: 9 (Graduate/Postgraduate)
Dates & Locations:

This subject has the following teaching availabilities in 2016:

Semester 2, Parkville - Taught on campus.
Pre-teaching Period Start not applicable
Teaching Period 25-Jul-2016 to 23-Oct-2016
Assessment Period End 18-Nov-2016
Last date to Self-Enrol 05-Aug-2016
Census Date 31-Aug-2016
Last date to Withdraw without fail 23-Sep-2016

Timetable can be viewed here. For information about these dates, click here.
Time Commitment: Contact Hours: 36 hours, comprising of two 1-hour lectures and one 1-hour workshop per week
Total Time Commitment:

200 hours


Admission into the 100 pt or 150 pt programs of the Master of Information of Technology


one of the following subjects:

Study Period Commencement:
Credit Points:
Semester 1, Semester 2
Semester 1, Semester 2


Recommended Background Knowledge:

Basic probability

Non Allowed Subjects:

433-484 Machine Learning
433-679 Evolutionary and Neural Computation
433-680 Machine Learning
433-684 Machine Learning

Core Participation Requirements:

For the purposes of considering request for Reasonable Adjustments under the Disability Standards for Education (Cwth 2005), and Student Support and Engagement Policy, academic requirements for this subject are articulated in the Subject Overview, Learning Outcomes, Assessment and Generic Skills sections of this entry.

It is University policy to take all reasonable steps to minimise the impact of disability upon academic study, and reasonable adjustments will be made to enhance a student's participation in the University's programs. Students who feel their disability may impact on meeting the requirements of this subject are encouraged to discuss this matter with a Faculty Student Adviser and Student Equity and Disability Support:


Dr Trevor Cohn



Subject Overview:


With exponential increases in the amount of data becoming available in fields such as finance and biology, and on the web, there is an ever-greater need for methods to detect interesting patterns in that data, and classify novel data points based on curated data sets. Learning techniques provide the means to perform this analysis automatically, and in doing so to enhance understanding of general processes or to predict future events.

Topics covered will include: supervised learning, semi-supervised and active learning, unsupervised learning, kernel methods, probabilistic graphical models, classifier combination, neural networks.

This subject is intended to introduce graduate students to machine learning though a mixture of theoretical methods and hands-on practical experience in applying those methods to real-world problems.


Topics covered will include: linear models, support vector machines, random forests, AdaBoost, stacking, query-by-committee, multiview learning, deep neural networks, un/directed probabilistic graphical models (Bayes nets and Markov random fields), hidden Markov models, principal components analysis, kernel methods.

Learning Outcomes:


On completion of this subject the student is expected to:

  1. Describe a range of machine learning algorithms
  2. Design, implement and evaluate learning systems to solve real-world problems, based on an appreciation of their relative suitability to different tasks
  • Two projects due around weeks 7 and 11, requiring approximately 65 - 70 hours of work in total (50%)
  • An end-of-semester examination not exceeding 3 hours (50%).

Hurdle requirement: To pass the subject, students must obtain:

  • A mark of at least 25/50 on the exam
  • and also a combined mark of at least 25/50 for the projects.

Assessment for this subject addresses both Intended Learning Outcomes (ILOs)

Prescribed Texts:


Breadth Options:

This subject is not available as a breadth subject.

Fees Information: Subject EFTSL, Level, Discipline & Census Date
Generic Skills:

On completion of the subject students should have the following skills:

  • Ability to undertake problem identification, formulation, and solution
  • Ability to utilise a systems approach to complex problems and to design and operational performance ability to manage information and documentation
  • Capacity for creativity and innovation
  • Ability to communicate effectively both with the engineering team and the community at large.


The subject is delivered through a combination of lectures and tutorials. One feature of the subject is that the projects are designed to be relatively open-ended and broad enough that students have scope to get hands-on experience implementing the breadth of material covered in the subject, as well as building off the subject content in innovating their own methods/researching related methods from the research literature and implementing them themselves.


Students will have access to lecture slides, readings relating to the lecture materials (both from a textbook and conference/journal papers), tutorial worksheets with worked solutions for all numeric problems, and sample reports to use in writing the project reports. Students are permitted to do their programming in any language and any programming environment/OS.


Machine learning has been growing rapidly in industry over the past two decades, with key industry players including Google, Microsoft, Amazon, Facebook and Twitter. There have been guest lecturers in the subject from organisations such as NICTA, which has a strong interest in machine learning (indeed one of the primary research groupings within NICTA is based on Machine Learning).

Related Course(s): Doctor of Philosophy - Engineering
Master of Information Technology
Master of Information Technology
Master of Philosophy - Engineering
Master of Science (Computer Science)
Related Majors/Minors/Specialisations: Approved Masters level subjects from other departments
B-ENG Software Engineering stream
MIT Computing Specialisation
MIT Distributed Computing Specialisation

Download PDF version.