Outline

Overview

The need to make sense of confusing, incomplete and noisy data is a problem central to virtually all branches of science. The underlying requirement is to draw robust, unbiased and insightful inferences from the data.After taking this course you should have a working knowledge of common data inference and model-fitting methods, and of machine learning techniques. You should be able to implement the model-fitting algorithms discussed here in your own code and use it to determine parameters from incomplete or noisy data. You will have a conceptual understanding of modern machine-learning techniques, including basic neural networks, and be able to implement your own network to solve a problem. Moreover, you will have the prerequisite knowledge to implement more complex machine learning architectures such as deep learning, using the wide range of available tools. The course is aimed to equip physicists (and other scientists) with practical tools to be deployed in their work, rather than delivering more theoretical content.

Unit details and rules

Academic unit	Physics Academic Operations
Credit points	6
Prerequisites ?	144 credit points of units of study including (12cp of MATH1XXX) or [(6cp of MATH1XXX) and DATA1X01]
Corequisites ?	None
Prohibitions ?	None
Assumed knowledge ?	48 credit points of 3000-level units of study and programming experience in Python
Available to study abroad and exchange students	Yes

Teaching staff

Coordinator	Peter Tuthill, peter.tuthill@sydney.edu.au
Lecturer(s)	Barnaby Norris, barnaby.norris@sydney.edu.au
Tutor(s)	Alison Wong, a.wong@sydney.edu.au

Type	Description	Weight	Due	Length
Final exam (Take-home short release)	Phys 4016 exam: Bayesian Data Inference and Machine Learning Written problems in Bayesian Statistics, MCMC and Machine Learning.	40%	Formal exam period	1.5 hours
Outcomes assessed:
Assignment	Assignment 1 2-3 solved problems with working and codes submitted.	12%	Week 05	3-4 pages
Outcomes assessed:
Assignment	Assignment 2 2-3 solved problems with working and codes submitted.	12%	Week 09	2-3 pages
Outcomes assessed:
Assignment	Assignment 3 2-3 solved problems with working and codes submitted.	12%	Week 13	2-3 pages
Outcomes assessed:
Tutorial quiz	Tutorial Completion Quiz Short submitted tutorial codes and solutions.	12%	Weekly	30 minutes.
Outcomes assessed:
Online task	Lecture Quizzes Brief quiz on weekly lectures, online, multiple-choice.	12%	Weekly	15 minutes
Outcomes assessed:
= Type D final exam ?

Assessment summary

Lecture Quiz: Content from each weeks lectures will be tested in the form of a short online quiz that probes the fundamental concepts for each week. This quiz should be completed, at the latest, prior to the Tutorial session each week. Total weight is 12% for all Lecture Quiz’s.

Tutorial Completion: Completion or substantial attempt at each week’s tutorial problems is rewarded with a total of 12% accumulated marks for the class.

Assignments: Three assignments, each contributing 12%, will consist of worked problems requiring mathematical and coding knowledge taken from the content taught in this class.

Final Exam: Students will sit a 90-minute final exam consisting of worked problems testing concepts in mathematical and coding algorithms central to this class.

Detailed information for each assessment can be found on Canvas

Assessment criteria

HD	High distinction	85 - 100	Awarded when you demonstrate the learning outcomes for the unit at an exceptional standard, as defined by grade descriptors or exemplars outlined by your faculty or school.
DI	Distinction	75 - 84	Awarded when you demonstrate the learning outcomes for the unit at a very high standard, as defined by grade descriptors or exemplars outlined by your faculty or school.
CR	Credit	65 - 74	Awarded when you demonstrate the learning outcomes for the unit at a good standard, as defined by grade descriptors or exemplars outlined by your faculty or school.
PS	Pass	50 - 64	Awarded when you demonstrate the learning outcomes for the unit at an acceptable standard, as defined by grade descriptors or exemplars outlined by your faculty or school.
FA	Fail	0 - 49	When you don’t meet the learning outcomes of the unit to a satisfactory standard.
AF	Absent fail	0 - 49	When you haven’t completed all assessment tasks or met the attendance requirements.

For more information see guide to grades.

Late submission

In accordance with University policy, these penalties apply when written work is submitted after 11:59pm on the due date:

Deduction of 5% of the maximum mark for each calendar day after the due date.
After ten calendar days late, a mark of zero will be awarded.

Academic integrity

The Current Student website provides information on academic integrity and the resources available to all students. The University expects students and staff to act ethically and honestly and will treat all allegations of academic integrity breaches seriously.

We use similarity detection software to detect potential instances of plagiarism or other forms of academic integrity breach. If such matches indicate evidence of plagiarism or other forms of academic integrity breaches, your teacher is required to report your work for further investigation.

Use of generative artificial intelligence (AI) and automated writing tools

You may only use generative AI and automated writing tools in assessment tasks if you are permitted to by your unit coordinator. If you do use these tools, you must acknowledge this in your work, either in a footnote or an acknowledgement section. The assessment instructions or unit outline will give guidance of the types of tools that are permitted and how the tools should be used.

Your final submitted work must be your own, original work. You must acknowledge any use of generative AI tools that have been used in the assessment, and any material that forms part of your submission must be appropriately referenced. For guidance on how to acknowledge the use of AI, please refer to the AI in Education Canvas site.

The unapproved use of these tools or unacknowledged use will be considered a breach of the Academic Integrity Policy and penalties may apply.

Studiosity is permitted unless otherwise indicated by the unit coordinator. The use of this service must be acknowledged in your submission as detailed on the Learning Hub’s Canvas page.

Outside assessment tasks, generative AI tools may be used to support your learning. The AI in Education Canvas site contains a number of productive ways that students are using AI to improve their learning.

Learning support

Simple extensions

If you encounter a problem submitting your work on time, you may be able to apply for an extension of five calendar days through a simple extension.  The application process will be different depending on the type of assessment and extensions cannot be granted for some assessment types like exams.

Special consideration

If exceptional circumstances mean you can’t complete an assessment, you need consideration for a longer period of time, or if you have essential commitments which impact your performance in an assessment, you may be eligible for special consideration or special arrangements.

Special consideration applications will not be affected by a simple extension application.

Using AI responsibly

Co-created with students, AI in Education includes lots of helpful examples of how students use generative AI tools to support their learning. It explains how generative AI works, the different tools available and how to use them responsibly and productively.

Weekly schedule

WK	Topic	Learning activity
Week 01	Introduction to Bayesian Reasoning	Independent study (1 hr)
	Bayesian Statistics: Fundamentals	Independent study (1 hr)
	Bayesian Fundamentals	Tutorial (2 hr)
Week 02	Parameter Estimation	Independent study (1 hr)
	Hypothesis Testing	Independent study (1 hr)
	Bayesian Computations	Tutorial (2 hr)
Week 03	Model Comparison: contrasting Bayes and Frequentist methods	Independent study (1 hr)
	Linear and Logistic Regression	Independent study (1 hr)
	Bayesian Testing	Tutorial (2 hr)
Week 04	Introduction to MCMC	Independent study (1 hr)
	MCMC: Tempering and Sampling	Independent study (1 hr)
	Building MCMC codes	Tutorial (2 hr)
Week 05	Advanced MCMC: Affine Invariant Samplers	Independent study (1 hr)
	Advanced MCMC: Hamiltonian Monte Carlo	Independent study (1 hr)
	Advanced MCMC	Tutorial (2 hr)
Week 06	Nested Sampling	Independent study (1 hr)
	Genetic Algorithms	Independent study (1 hr)
	Advanced numerical methods	Tutorial (2 hr)
Week 07	Resampling and Approximate Bayesian Computation	Independent study (1 hr)
	Introduction: Information and Entropy	Independent study (1 hr)
	Resampling, Information, Entropy	Tutorial (2 hr)
Week 08	Information Criteria and Hierarchical Bayes	Independent study (1 hr)
	Inverse Problems and Image Recovery	Independent study (1 hr)
	Advanced Bayes & inverse problems	Tutorial (2 hr)
Week 09	Compressed Sensing	Independent study (1 hr)
	Gaussian Processes	Independent study (1 hr)
	Data dimensionality and Gaussian Processes	Tutorial (2 hr)
Week 10	Introduction to machine learning	Independent study (1 hr)
	Dimensionality reduction and basic supervised learning	Independent study (1 hr)
	Introduction to Machine Learning	Tutorial (2 hr)
Week 11	Introduction to neural networks	Independent study (1 hr)
	Deep learning with neural networks	Independent study (1 hr)
	Introduction to Neural Networks	Tutorial (2 hr)
Week 12	Deep learning architectures and applications for inference	Independent study (1 hr)
	Generative and Bayesian neural networks	Independent study (1 hr)
	Deep Learning	Tutorial (2 hr)
Week 13	Summary and Revision	Independent study (1 hr)
	Summary and Revision	Independent study (1 hr)
	Summary and revision	Tutorial (2 hr)

Study commitment

Typically, there is a minimum expectation of 1.5-2 hours of student effort per week per credit point for units of study offered over a full semester. For a 6 credit point unit, this equates to roughly 120-150 hours of student effort in total.

Learning outcomes

Learning outcomes are what students know, understand and are able to do on completion of a unit of study. They are aligned with the University's graduate qualities and are assessed as part of the curriculum.

Outcomes
Graduate qualities

At the completion of this unit, you should be able to:

LO1. Understand the guiding philosophy of the Bayesian approach to probability and its application to data processing and parameter es�ma�on, contrasting this to a frequentist approach.
LO2. Apply Bayesian principles in data analysis, specifically in the roles of hypothesis testing and parameter es�timation.
LO3. Have a fundamental understanding of the very distinct ways in which Bayesian statistics differs from much of 20th century practice (Frequentist Statistics), and in particular in the domain of how hypothesis testing is framed and conducted.
LO4. Understand the meaning and application of Maximum-Entropy techniques in determining maximally ignorant probability distributions, and the extraction of information through image deconvolution.
LO5. The ability to apply these concepts to develop statistical and machine learning models, and to solve qualitative and quantitative problems in scientific and engineering contexts, using appropriate mathematical and computing techniques as necessary.
LO6. Understand the challenge posed by incomplete and noisy data, and the importance of using robust tools to infer underlying information and reliably quantify uncertainty in inferences or predictions.
LO7. Be able to apply inference and model-fitting tools (such as Markov chain Monte Carlo) to train models on real data, and have a fundamental understanding and intuition for how these tools work, and their strengths and weaknesses.
LO8. Understand the utility of unsupervised machine learning and how it contrasts with traditional approaches to categorisation and inference, and be able to recognise its dangers and limitations.
LO9. Understand the fundamental principles of neural networks and deep learning: be able to recognise the types of problems that are suited to such techniques, and appreciate their power as compared to previous approaches.
LO10. Be able to apply neural network based machine-learning techniques to actual data sets to perform useful data inference, and be prepared for implementing more complex machine learning models using the wide variety of available frameworks.

Graduate qualities

The graduate qualities are the qualities and skills that all University of Sydney graduates must demonstrate on successful completion of an award course. As a future Sydney graduate, the set of qualities have been designed to equip you for the contemporary world.

GQ1	Depth of disciplinary expertise Deep disciplinary expertise is the ability to integrate and rigorously apply knowledge, understanding and skills of a recognised discipline defined by scholarly activity, as well as familiarity with evolving practice of the discipline.
GQ2	Critical thinking and problem solving Critical thinking and problem solving are the questioning of ideas, evidence and assumptions in order to propose and evaluate hypotheses or alternative arguments before formulating a conclusion or a solution to an identified problem.
GQ3	Oral and written communication Effective communication, in both oral and written form, is the clear exchange of meaning in a manner that is appropriate to audience and context.
GQ4	Information and digital literacy Information and digital literacy is the ability to locate, interpret, evaluate, manage, adapt, integrate, create and convey information using appropriate resources, tools and strategies.
GQ5	Inventiveness Generating novel ideas and solutions.
GQ6	Cultural competence Cultural Competence is the ability to actively, ethically, respectfully, and successfully engage across and between cultures. In the Australian context, this includes and celebrates Aboriginal and Torres Strait Islander cultures, knowledge systems, and a mature understanding of contemporary issues.
GQ7	Interdisciplinary effectiveness Interdisciplinary effectiveness is the integration and synthesis of multiple viewpoints and practices, working effectively across disciplinary boundaries.
GQ8	Integrated professional, ethical, and personal identity An integrated professional, ethical and personal identity is understanding the interaction between one’s personal and professional selves in an ethical context.
GQ9	Influence Engaging others in a process, idea or vision.

Outcome map

Learning outcomes	Graduate qualities
Learning outcomes

Responding to student feedback

This section outlines changes made to this unit following staff and student reviews.

First time unit offered.

Disclaimer

The University reserves the right to amend units of study or no longer offer certain units, including where there are low enrolment numbers.

To help you understand common terms that we use at the University, we offer an online glossary.

Current students

Overview

Overview

Unit details and rules

Teaching staff

Assessment

Assessment

Assessment summary

Assessment criteria

Late submission

Academic integrity

Learning support

Learning support

Simple extensions

Special consideration

Using AI responsibly

Weekly schedule

Weekly schedule

Study commitment

Learning outcomes

Learning outcomes

Graduate qualities

Outcome map

Responding to student feedback

Responding to student feedback

Disclaimer

On this page

Useful links

Media

Student links

About us

Connect

Current students

PHYS4016: Bayesian Data Inference and Machine Learning

Semester 2, 2021 [Normal day] - Remote

Overview

Overview

Unit details and rules

Teaching staff

Assessment

Assessment

Assessment summary

Assessment criteria

Late submission

Academic integrity

Learning support

Learning support

Simple extensions

Special consideration

Using AI responsibly

Weekly schedule

Weekly schedule

Study commitment

Learning outcomes

Learning outcomes

Graduate qualities

Outcome map

Responding to student feedback

Responding to student feedback

Disclaimer

On this page

Useful links

Media

Student links

About us

Connect