1. Handbook
  2. Subjects
  3. Trustworthy Machine Learning

Trustworthy Machine Learning (COMP90073)

Graduate courseworkPoints: 12.5On Campus (Parkville)

You’re viewing the 2026 Handbook:
Or view archived Handbooks

Overview

Availability
Semester 2 - On Campus
FeesLook up fees

AIMS

As machine learning systems are increasingly integrated into critical and data sensitive applications, ensuring their confidentiality, reliability, robustness, and fairness becomes imperative. The complexity of modern AI models, coupled with evolving threats such as data inference, adversarial attacks, data poisoning, and biases, necessitates new methodologies to build and evaluate trustworthy machine learning systems. Trustworthy Machine Learning will explore techniques to enhance the privacy, security, interpretability, and safety in deployment of machine learning models, ensuring they operate reliably in real-world environments. 

INDICATIVE CONTENT

The subject will begin by introducing the key dimensions of trustworthiness in machine learning, including privacy, robustness, reliability, and fairness. Students will examine real-world case studies that highlight failures and vulnerabilities in deployed AI systems.

The first part of the subject will explore different types of information leakage that can arise under various threat models and examine methods for protecting sensitive data during analysis. The second part will introduce machine learning techniques that enhance model reliability, with a particular focus on unsupervised learning methods such as anomaly detection, alarm correlation, and intrusion detection. The third part of the subject will introduce some of the theoretical challenges and emerging issues for security analytics research, based on recent trends in the evolution of security threats. 

By the end of the subject, students will gain both theoretical knowledge and practical skills to design, evaluate, and deploy machine learning models with trustworthiness as a core principle. 

Intended learning outcomes

On completion of the subject, students should be able to:

  • Identify and evaluate vulnerabilities in machine learning systems, including risks related to privacy, adversarial attacks, data poisoning, and biases.
  • Describe and implement techniques for enhancing the trustworthiness of machine learning models, including privacy-preserving methods, robustness enhancements, and fairness-aware learning.
  • Analyse different types of information leakage in machine learning models and apply methods to mitigate such risks.
  • Implement and apply anomaly detection, alarm correlation, and intrusion detection techniques to improve security and reliability in machine learning applications.
  • Assess the reliability and robustness of machine learning models in adversarial environments, including their susceptibility to adversarial and poisoning attacks.
  • Evaluate the effectiveness of methodologies for enhancing trustworthiness based on accuracy, efficiency, and real-world applicability.
  • Discuss emerging challenges and research trends in securing and ensuring the privacy and integrity of machine learning systems.

Generic skills

  • Ability to undertake problem identification, formulation and solution
  • Ability to utilise a systems approach to complex problems
  • Capacity for creativity and innovation
  • Ability to communicate the results of complex analysis effectively to both technical audiences and the community at large

Last updated: 13 March 2026