CDMS Research Start-up Fund Projects 2024
Project Leader: Dr Shatha Aziz
Project Title: Neural Differential Equations: Solutions and Mathematical Modelling
Research team: Mentors: Professor Paul Hurley and Associate Professor Yi Guo
Project Summary:
Differential Equations (DEs) plays a crucial role in modelling many real-life problems such as problems related to financial mathematics, epidemiology, ground-water level, population growth, advection-diffusion and gas dynamics. Numerical, approximated and optimisation techniques are used to solve these types of equations. Interdisciplinary research provides an opportunity to achieve remarkable outcomes for both disciplines. There are existing similarities between deep neural networks and DEs. With the right choice of activation function, we can get smooth first and second derivatives of the nonlinear function which represent these neural networks. Thereby these types of deep neural networks can be used to solve complex differential equations. In addition to this, neural differential equations can be used to automate mathematical modelling process and model problems which don’t have certain structure using a given data set. These neural DEs require less memory use, are adaptive and can increase the solution accuracy of complex problems. This project aims to solve differential equations using deep learning techniques and explore automated mathematical modelling approaches. The project will contribute to filling some existing gaps and to the body of the literature in deep learning and DEs fields.
Project Leader: Zhonglin Qu
project title: Improving trust in medical domain decision-making with explainable artificial intelligence and data storytelling
Research team: Quang Vinh Nguyen
Summary of your project:
Trust in artificial intelligence (AI) is getting critical, especially in the medical domain. Medical domain practitioners require more detailed information about the mechanisms of AI black box decisions to take action based on the computational decisions trustfully. Human-computer interactions (HCI) practitioners try to use visual analytics techniques, such as high accuracy models, AI transparency, communication uncertainty, showing provenance, expressiveness, providing multiview, adding interactions, and adding social factors to improve AI trustworthiness to enhance understanding and inspire complete trust in the machine learning models.
Data storytelling conveys insights from data in a narrative format, making complex information more understandable and compelling to a broader audience. Data storytelling involves structuring the data into a coherent narrative that resonates with the audience, often using visualisations, anecdotes, and real-world examples to illustrate key points. Data storytelling leverages the art of visuals and narratives to make data communication understandable, engaging, and enlightening in a trustworthy way, ultimately driving changes.
This research combines explainable AI and data storytelling in the medical data analysis domain and may extend to the indigenous data analysis domain. The study involves reviewing AI evaluation to identify the current explainable AI (XAI) gaps, soliciting domain users’ input, and developing novel XAI models.
Project Leader: Dr Rhys Tague
Research team: Assoc Prof Anupama Ginige
Project Title: Enhancing Formative Assessment in Computer Capstone Subjects with Generative AI
Project Summary: The capstone subject for computing degrees at Western Sydney University (WSU), known as Professional Experience, involves students solving complex problems under academic supervision and in consultation with industry clients. Teams of 3 to 4 final-year students work on unique projects that reflect client needs. Each week, students consult with clients, academic supervisors, and team members to develop a project that meets client requirements. This approach mimics real-world scenarios, fostering holistic learning and applying theoretical concepts.
Given the challenges of this self-directed learning model, two information sessions help students understand the structure. Despite formative learning opportunities, feedback varies between supervisors and is group-focused. Therefore, an additional resource for self-reflection and personalised feedback, such as the proposed Personal AI Capstone Mentor, could enhance student approaches to assessments and team dynamics.
The hypothesis is that integrating GenAI for personalised feedback will improve self-assessment, team dynamics, and learning outcomes in capstone subjects. The project will develop the AI mentor, integrate it into the subject, and evaluate its effectiveness through trials. This innovative tool aims to provide scalable, personalised feedback, contributing to new knowledge in educational pedagogy, reflective learning, and team dynamics.
Project Leader: Mengfan Lyu
Project title: Generalised Temperley-Lieb algebras
Research team: Dr. Mengfan Lyu and A/Professor James East(mentor)
Project summary:
Classical Temperley-Lieb algebras play a vital role in various areas of mathematics, including knot theory and quantum mechanics. However, the knowledge about the Temperley-Lieb algebras corresponding to complex reflection groups is still limited. This project addresses this gap by introducing generalized TL algebras (GTLAs) that encompass the classical types as special cases. This broader framework allows for the potential application of TL algebras to a significantly wider range of reflection groups. The research questions focus on exploring the properties of GTLAs, including their cellularity, potential diagrammatic realization, and connections to other areas of mathematics. The research is expected to yield significant new knowledge by extending the theory of Temperley-Lieb algebras and their applications. The development of generalized algebras and their cellular structures will provide new tools and insights for mathematicians and physicists, advancing both theoretical understanding and practical applications.
Project Leader: Mehdi Tavakol
Research team: Roozbeh Hazrat, Julien Ugon (Deakin University)
Project title: Applying Algebraic Geometry to Quantum Computing
Project summary:
The current project focuses on applying methods from algebraic geometry to quantum computing by studying the isogeny graph associated with supersingular curves and analyzing the action of the ideal class group. This recent approach aims to deepen our understanding of the structural properties of isogeny graphs and their relevance in cryptographic protocols resistant to quantum attacks. By using methods from algebraic geometry, the project seeks to address challenges such as efficiently computing isogenies between supersingular elliptic curves, thus enhancing the practical viability of cryptographic schemes in the post-quantum era. By exploring the interplay between algebraic geometry and quantum computing, the project aims to contribute to the development of novel cryptographic primitives with improved security and efficiency.
Project Leader: Chng Wei LAU
Project Title: Uncovering Disparity in Cancer care patients: data visualisation approach
Research Team: Madhushi Bandara Madhushi.Bandara@uts.edu.au (at UTS), Oliver Oobst
Project summary:
The majority of research on cancer patient care pathways is international and limited in scope. Current data mining techniques have limitations, including data oversimplification, lack of generalizability, and difficulty handling complex clinical processes. A study focusing on Australian cancer patients could improve local outcomes and address care variations that negatively impact patients. This project aims to discover new temporal patterns and develop visualization methods using MIMIC-IV and Synthea synthetic data, adapting previously tested methodologies. MIMIC-IV is a publicly available electronic health record database, and Synthea generates synthetic patient data. These data require transformation into the OHDSI OMOP Common Data Model for research. The project will validate analysis methods with Synthea data before applying them to MIMIC-IV and actual Australian patient data. The project involves converting MIMIC-IV data to OMOP CDM and transforming the OMOP data into an event-driven format for temporal pattern analysis and visualization using graph theory.