Prof Jin-Song Dong

This project aims to develop safer and more secure AI decision-making systems for the medical domain. We plan to develop a new learning approach which combines probabilistic model checking and reinforcement learning and provides formal safety guarantees for the learned policies. This learning approach will be integrated into an adversarial learning framework which trains a target agent and an adversarial agent simultaneously. The goal is to make the target agent "immune" to adversarial attacks, thus improving the security of the system. Finally, we will apply our method to Datarwe's medical applications such as COVID-19 respirator and drug dosage related decision-making.

Prof Johnson Mak

Respiratory infections and antibiotic resistance bacterial infection are some of the conditions that have significantly stressed our hospital ICU. A number of risk factors have been reported to associate with severe diseases, which includes age, pre-existing conditions, pathogen setpoints, responsiveness to therapeutic strategy. Any single risk factor is unlikely to be an absolute determinate of clinical diseases, rather many contributors or associations are important with the disease progress. We will use ICU data and artificial intelligence to generate a prediction algorithm to assist clinical decision making.

Dr Ping Zhang

This project aims to develop a real time application for prediction of intensive care outcomes in traumatic brain injuries. It utilises heart rate variability parameters, applies ECG signal analysis techniques in combination with machine learning and feature selection algorithms. The candidate will investigate time series analysis and various classification methodologies, and will develop a software pipeline from data collection to prediction model. Experience in computer programming (eg. Python) is required. Basic knowledge in statistics, time series signal analysis, artificial neural network and genetic algorithm are preferred.

Dr Zhe Wang

The research project focuses on developing explainable AI solutions to decision support systems, by combining knowledge graphs, machine reasoning and machine learning. Knowledge graphs is a promising data and knowledge organisation, synthesis and management approach, and we have developed scalable reasoning tools for knowledge graphs coupled with ontological rules that describe domain knowledge or business rules. Therefore, this project aims to study the problem of incorporating such high-level knowledge and formal reasoning in the analysis of cross-media data (obtained from vision, sound, language and other sources). Moreover, such knowledge and reasoning can be integrated with machine learning models to provide powerful support for informed decision-making where a justification or explanation of the decision can potentially be retrieved

Assoc Prof Alan Wee-Chung Liew

This project aims to develop novel stream learning algorithms for continuous patient outcome prognosis by taking into account patient's data collected during ICU admission in a unified manner. The algorithms are expected to integrate high frequency time series data with patient's demographic  data, lab data, diagnosis data, prescription data, etc. as exemplified in MIMIC-III, for accurate outcome prognosis. Issues such as prediction bias, data leakage, data sparsity, non-stationarity, model explainability will be investigated.

Supervisors: A/Prof V. Muthukkumarasamy, Dr Zhe Hou

Description: Blockchain is a promising technology towards achieving full-scale digital transformation in a complex environment. This technique has attracted a number of successful applications such as cryptocurrency, supply chain, trade finance and smart contracts.  This peer-to-peer technology is adopted across a range of industry sectors including manufacturing, agriculture, finance, health and e-governance. For example, smart contracts facilitate real-time transactions across multiple jurisdictions with untrusted parties, while providing provenance and integrity. Blockchain has been show-cased as a game changing national technological strategy in several countries.One of the major research problems is about how to autonomously and safely operate and transfer digital assets amidst potential adversarial entities. The research challenge includes integration of different trust mechanisms, guaranteeing security properties, and providing scalability.This project will extend our current work on the classification of digital assets, cross-chain integration protocols, and formal verification of smart contracts with novel design patterns and formal security guarantees for inter-blockchain systems. Experiments and validation will be carried out on test-networks and using real-world case studies with our industry collaborators

Supervisors: Prof Jin Song Dong, Prof Yongsheng Gao, Dr Zhe Hou, Dr Hadrien Bride

Description: Machine learning has been used in data analytics for insurance, sports, tourism, marketing and many other areas. However, most existing machine learning algorithms often give excellent prediction results without telling the user how the decisions are made. This weakness results in trust issues from the user and limitations for adopting machine learning in some applications. In recent years, there have been many accidents caused by not understanding the trained model before deploying it. This project aims to leverage the fusion of machine learning and automated reasoning and develop next-generation machine learning methods that provide not only excellent predictions but also user-friendly explanations as well as formal guarantees of the correctness and safety of prediction models.

Supervisors: Dr. Hui Tian and A/Prof. Alan Liew

Research Abstract: With the exponential growth of streaming data from various sources in both volume and content, privacy protection for streaming data and their secure analysis are becoming increasingly important. Considering the properties of streaming data including mass volume (unbounded size), heterogeneity, dynamicity, concept drift and feature evolution  this project applies multi-fold theories and techniques including secure computing, privacy protection, machine learning, intelligent searching, data mining  in an effectively coordinated way. The project first studies how to discover and measure sensitive information of data instances, including features and labels, in data streams. It then investigates suitable models, schemes and mechanisms for effective protection of the sensitive information while preserving the required data utility. Finally it develops new techniques and methods for various privacy-preserving streaming data analysis and mining, including statistical analysis, association mining, classification and clustering, and evaluation of their performance.

Supervisors: Prof. Vladimir Estivill-Castro and Prof. Abdul Sattar.

Description: There is significant excitement about the recent advances in artificial intelligence because of applications of deep learning, task planning, and reinforcement learning resulting in super-human performance for some tasks. However, in the case where agents must make decisions holding incomplete information cooperating, but with teammates and facing adversaries, there is no clear answer to how to integrate reinforcement learning with deep learning to create optimal or effective policies. That is, while the situation can be formally represented as a partially observable Markov decision process (POMDP), the combinatorial explosion of states and the possibility to include Bayesian inference or other reasoning leaves open the construction of effective artificial agents. The challenge is to design agents and find integrated architectures in several test cases.

Supervisors: Prof Kewen Wang and Dr Zhe Wang

Summary: Intelligent systems are important for cross-media (e.g., text, image, video, and audio) retrieval, web content monitoring, web information trend analysis, and healthcare data fusion. The state-of-the-art AI approaches to cross-media analysis mostly adopt a black-box neural networks approach, which cannot provide human-understandable explanations for specific decisions made. This research aims to develop explainable AI techniques for analysis and reasoning of cross-media contents by integrating deep learning with symbolic knowledge representation. This project is based on the unique strength of the supervisors' research group and is a critical part of a proposed future ARC project.

We seek strong PhD candidates with a good background in Computer Science/IT, Mathematics, Engineering or Philosophy.

Supervisors: Dr Henry Nguyen (Principal), Associate Professor Jun Jo (Associate)

Description: Our modern society is struggling with an unprecedented amount of online fake news, which do harm to democracy, economics, and national security. Creators of fake news optimise their chance to manipulate public opinion and maximise their financial and political gains through sophisticated pollution of our information diffusion channels. Such attacks are driven by the advances of modern artificial intelligence these days and pose a new and ever-evolving cyber threat operating at the information level, which is far more advanced than traditional cybersecurity attacks at the hardware and software levels. In this project, we will investigate several research challenges, including deep understanding of the spread of fake news online, early detection methods, and mitigation strategies. This project is linked to the current ARC DECRA and can be submitted to other ARC grant application/project in the future.

This project aims at developing novel techniques and tools for the automatic detection of anomalies based on times-series data from critical care units collected at the PMDP platform, with the eventual goal of building realistic tools that can aide medical practitioners and improve patient care. We will explore existing time-series data analysis methods (e.g., traditional statistical tools, LSTM-networks, deep neural networks) and create new models and algorithms that suit our purpose. From a technical perspective, we are especially interested in novel graph-based sequential pattern detection techniques in time-series data that can build on our existing work on graph analysis

Pre-existing research has failed to offer a solution to protect patients’ privacy and confidentiality, it is important to identify the limitations of existing solutions and envision directions for future research in privacy preservation in health informatics. This research aims to identify current outstanding issues that act as impediments to the successful implementation of privacy measures in health informatics and the limitations of available solutions. Feasibility of using blockchains for dealing with health and medical will be researched and evaluated. Then, propose a privacy-preserving framework by improving data storage, record linkage techniques.

There integration of pathological analysis across time and across population can provide longitudinal data for data analysis based on machine learning techniques. The goal is to anticipate and predict the needs of patients to have confirmatory pathological studies, or to optimise the tracking of the evolution of their health. Data analysis should enable optimise the coast of studies but maximise the opportunity to detect variations on the patient’s condition. We aim to reduce the need of an expensive (costly or/and invasive) medical tests by creating algorithms suggesting the best opportunity for therapy and the most suitable test to monitor the evolutions of the patient.