Directed Studies Projects and Supervisors
Hallucination-Resistant Large Language Models for Engineering Systems
Project Supervisor: Tansu Alpcan
Offered: Semester 1 and Semester 2
Large language models (LLMs) have emerged as powerful artificial intelligence tools with widespread applications. However, their unreliability and propensity for hallucination present significant challenges and constitute a major barrier to their adoption in engineering systems. This project aims to address these challenges by developing a principled framework for evaluating the consistency and reliability of LLMs as decision-support tools in engineering contexts.
Detecting hallucinations is inherently challenging, as it typically requires knowledge of the effects of actions, which is often unavailable. An alternative approach is to assess the internal consistency of actions generated by the LLM and to interpret inconsistencies as indicators of hallucination. From a logical perspective, an action is considered inconsistent if it contradicts itself. Building on this logical foundation, the project will develop metrics to quantify action consistency by identifying contradictions among multiple candidate actions. Intuitively, a higher degree of contradiction among candidate actions suggests an increased risk of hallucination.
The project will investigate methodologies such as integrating the LLM within an iterative feedback loop, in which candidate actions are generated and subsequently evaluated for consistency with system constraints and lookahead predictions. In addition, system designs that enable control of hallucination risk through adjustable consistency thresholds and in-context learningwill be explored, and performance metrics such as regret bounds will be examined. The proposed approaches will be evaluated through their application to engineering decision-support problems in communication networks and/or security systems.
Overall, the project will combine theoretical analysis with computational implementation. Interested students should possess strong backgrounds in engineering and computing, as well as solid analytical and programming skills.
Extension of Phase and Gain Margins to MIMO Systems with Application to Power Networks
Project Supervisor: Farhad Farokhi
Offered: Semester 1 and Semester 2
This summer research project investigates the characterization and analysis of multi-input multi-output (MIMO) phase and gain margins for linear systems, with a particular focus on their application to modern power systems. Traditional single-input single-output (SISO) stability margins provide limited insight into the robustness of interconnected and high-dimensional systems, such as power grids with multiple generators and loads. By extending classical robustness concepts to the MIMO setting, this project aims to derive tractable measures of phase and gain margins that capture cross-coupling effects between system channels. The theory will then be applied to linearised models of power system dynamics to assess how variations in control gains, line impedances, and communication delays affect overall system robustness.
Optically controlled 3D micro-robots
Project Supervisor: Daniel Fan
Offered: Semester 1 and Semester 2
Soft materials based micro-mechanisms can be manipulated using optical or electronic traps, allowing interactions with micro-objects such as cells. One interesting design strategy that will be explored in this project is the use of kirigami and origami to fold 2D planar structures into 3D shapes. In this way, complicated mechanisms such as force actuators, adaptive micro-optics, and active micro-fluidic components can be realised.
In this project, the student will program an existing microscope setup to produce arbitrary light patterns suitable for micro-robot actuation. Then, they will design a simple 3D micro-robot that performs out-of-plane motion, which will be fabricated at the Melbourne Centre for Nanofabrication. Finally, the micro-robot will be experimentally characterised using the programmed light patterns. Such soft-micro-robotics can be used for interactions in liquid environments such as focusing a micro-lens, pulling/pushing on a cell, or controlling micro-fluidic flows via valves, pumps, and gates.
Transparent transistors
Project Supervisor: James Bullock
Offered: Semester 1 and Semester 2
Transparent oxide electronics, such as those based on indium and zinc oxides, are exciting emerging technology platforms. They can be deposited at low temperatures on flexible/plastic substrates enabling various applications such as augmented reality displays and other wearable electronics.
This Directed Studies project will explore the fabrication and characterization of oxide-based field effect transistors on transparent substrates. It will involve work in a laboratory environment to fabricate devices and is an excellent project for those considering a career in electronics research. It will also involve measuring and modelling the behavior of fabricated devices to identify their potential and opportunities for further optimization. This project would suit someone in the electronics pathway of the EEE Masters course.
Tandem solar cell advanced characterisation
In many parts of the world, solar cells have become the cheapest source of electricity. The next generation of industrial solar cells is expected to use tandem architectures, where two or more cells are stacked to better extract energy from the solar spectrum. The most promising design combines a top-cell made from a new semiconductor that absorbs light up to about 1.7 eV, with a silicon bottom cell that absorbs light down to 1.1 eV. Developing suitable top-cell materials requires new fabrication and characterization techniques.
When optically or electrically excited, solar-cell materials emit photons with energies corresponding to their bandgaps in a process known as photoluminescence or electroluminescence. The intensity of this emission is proportional to the concentration of excess electrons and holes generated in the material. Measuring the intensity and spectrum of this emitted light provides valuable insight into the optoelectronic quality of the device.
This Directed Studies project will include both practical and theoretical components. The practical work will focus on measuring photoluminescence images of large-area top-cell materials, while the theoretical work will involve developing models to convert the measured photoluminescence into estimates of the implied device voltage.