• Offered by Research School of Computer Science
• ANU College ANU College of Engineering and Computer Science
• Course subject Computer Science
• Mode of delivery In Person
• Co-taught Course
• Offered in
Advanced Topics in Statistical Machine Learning (COMP4680)

This course explores a selected area relevant to statistical machine learning in depth, and will be taught by an SML staff member of internationally recognised standing and research interest in that area. Based on current SML staffing, this will be one of:

•    kernel methods
•    graphical models
•    reinforcement learning
•    convex analysis
•    optimisation
•    bioinformatics
•    minimal description length principle
•    topics in information theory
•    decision theory

Over the past several years the content has alternated between “convex analysis and optimisation” and “structured probabilistic models”. Students should contact the course convenor to find out what topic is planned for the coming semester.

## Learning Outcomes

Upon successful completion, students will have the knowledge and skills to:

The learning outcomes change based on the area taught. For the convex analysis and optimisation topic, at the end of the course students should be able to:
•  Distinguish definitions of key concepts in convex analysis, including convexity of sets and functions, subgradients, and the convex dual
•  derive basic results about convex functions such as Jensen’s inequality
•  Deduce how Bregman divergences are constructed from convex functions and derive some of their properties
•  Produce a formal optimization problem from a high-level description and determine whether the problem is convex
•  Recognize standard convex optimization problems such as linear programs and quadratic programs
•  Derive the standard (dual) quadratic program for support vector machines and understand the extension to max-margin methods for structured prediction
•  Implement and analyse gradient descent algorithms such as stochastic gradient descent and mirror descent
For the structured probabilistic models topic, at the end of the course the student should be able to:
• Distinguish definitions of key concepts in probabilistic graphical models, including Bayesian networks, Markov networks, probabilistic queries, and conditional independence
• Derive independence assumptions from graphical representations of a probabilistic model
• Contrast and implement various exact and approximate inference algorithms, including belief propagation of trees, sampling, and variational schemes
• Derive maximum likelihood learning for probabilistic graphical models
• Demonstrate advanced understanding of approximations of the likelihood function

## Indicative Assessment

Assessment will be in the form of regular assignments and an open-book final examination.

The ANU uses Turnitin to enhance student citation and referencing techniques, and to assess assignment submissions as a component of the University's approach to managing Academic Integrity. While the use of Turnitin is not mandatory, the ANU highly recommends Turnitin is used by both teaching staff and students. For additional information regarding Turnitin please visit the ANU Online website.

2hrs lectures weekly 8 hours of self study per week

## Requisite and Incompatibility

To enrol in this course you must have completed COMP4670. Incompatible with COMP8650.

## Prescribed Texts

Main text (depending on the topic taught):

• Stephen Boyd and Lieven Vandenberghe, "Convex Optimization" (convex optimisation)
• Kevin Murphy, "Machine Learning" (structured probabilistic models)

Reference texts:

• Hiriart-Urruty and Lemaréchal, “Fundamentals of Convex Analysis”
• Bertsekas, Nedic and Ozdaglar, “Convex Analysis and Optimization”
• Bertsekas, “Nonlinear Programming”
• Koller and Friedman, "Probabilistic Graphical Models"
• Bishop, "Pattern Recognition and Machine Learning"

## Assumed Knowledge

• Knowledge of machine learning at the level of COMP4670 Introduction to SML
• Familiarity with linear algebra (including norms, inner products, determinants, eigenvalues, eigenvectors, and singular value decomposition)
• Familiarity with basic probablity theory
• Familiarity with multivariate differential calculus (e.g., derivative of a vector-valued function)
• Exposure to mathematical proofs

## Fees

Tuition fees are for the academic year indicated at the top of the page.

If you are a domestic graduate coursework or international student you will be required to pay tuition fees. Tuition fees are indexed annually. Further information for domestic and international students about tuition and other fees can be found at Fees.

Student Contribution Band:
2
Unit value:
6 units

If you are an undergraduate student and have been offered a Commonwealth supported place, your fees are set by the Australian Government for each course. At ANU 1 EFTSL is 48 units (normally 8 x 6-unit courses). You can find your student contribution amount for each course at Fees.  Where there is a unit range displayed for this course, not all unit options below may be available.

Units EFTSL
6.00 0.12500

## Course fees

Domestic fee paying students
Year Fee
2017 \$3660
International fee paying students
Year Fee
2017 \$4878
Note: Please note that fee information is for current year only.

## Offerings, Dates and Class Summary Links

The list of offerings for future years is indicative only.
Class summaries, if available, can be accessed by clicking on the View link for the relevant class number.

### Second Semester

Class number Class start date Last day to enrol Census date Class end date Mode Of Delivery Class Summary
9130 23 Jul 2018 30 Jul 2018 31 Aug 2018 26 Oct 2018 In Person N/A