Hierarchical Planning is a powerful formalism to express control rules on how plans may be generated. Here, action sequences are generated from an initially given abstract task, which is step-wise refined until an executable primitive action sequence is obtained (similar to formal grammars/languages). Hierarchical Planning is so powerful that it allows to express undecidable problems -- however in practice most problems are decidable as special cases allow for (terminating) decision procedures. Many special cases are already known, yet several more remain unexplored. This project is concerned with exploring such open special cases.

The exact problem to be investigated is still to be decided upon. We could either investigate further classes of standard hierarchical planning or look into yet completely unexplored territories such as hierarchical planning with time -- to name just two options.

This is a highly challenging project, where you should only apply to if you are deeply interested in theory. You *must* bring with a decent understanding of complexity theory as you will have to conduct proofs by yourself. Best case: You completed the course COMP3630 or COMP6363 (Theory of Computation). If in doubt, drop me an email.

Candidate will be developing case studies of the use of model checking (and/or Symbolic AI) technology in the application of developing simulation/controller codes. Workflows in this domain are currently undertaken using detailed simulation studies that appeal to hard/expensive to acquire data. These expensive incumbent workflows identify potential issues with a simulator/controller, which can then be addressed by scientists and engineers. Codes are developed over years, an become quite complex and thereby potentially error prone. Model checking and symbolic AI are motivated here, to push the envelope on the class of errors which can be eliminated prior to expensive computational or physical resources are expended.

Experimental approach

This project contemplates the development and empirical study of simplistic codes, derived by the candidate should they wish to do that. Many codes (Fortran and C++) that are amenable to model checking in practice are already available for study.

Possible prospective investigations

For a class of potential issues—e.g., (1) exceeding superconducting coil current limits, or (2) simulated soundwave in a plasma exceeding the speed of light—determine whether existing software model checking tools are able to efficiently reason about those issues, and characterise the computational effort of doing so. 

General information about the Work Group, the University and the region

Collaborations between Australian and European scientists played a pioneering role in fusion science, with the first laboratory fusion reaction realised by Mark Oliphant (Australian) and Paul Harteck (Austrian). The tradition of Oliphant’s pioneering research continues at the Australian National University (ANU), which is ranked 1st in Australia, and 27th in the world, by the QS World University Rankings, 2022, and 2nd in Australia, and 54th in the world, by Times Higher Education Rankings, 2022. Australia has a 77 year history of studying plasma physics, with ANU playing a historical and central role, being for example the first institution in the world outside the (then) USSR to host experimental Tokamaks (1964-69).  With regulatory changes in April 2023 in the USA, opening the door to an significant 21st century commercial fusion industry, now is an exciting time to get involved in fusion science and the development of the enabling AI technologies for that. 

The student will be working at the ANU under the supervision of computer scientist Dr Charles Gretton. Dr Gretton is a member of the Intelligence Cluster in the School of Computing, which is part of the College of Engineering and Computer Science at the Australian National University. Dr Gretton collaborates with and is co-located with fusion science experts, such as Prof. Matthew Hole in the Mathematical Sciences Institute.

Canberra is a sunny modern city located on the traditional lands of the Ngunawal and Ngambri peoples, who lived in the region for more than 20,000 years. The city is the so-called “bush capital", showcasing the unique flora and fauna adjacent Australia’s eastern seaboard. Canberra is the capital city of Australia, located ~3.5 hours from Sydney by road, and 1 hour from Sydney by plane. In Canberra you are also a short (bus) drive from beautiful beaches (Batemans Bay), and from the highest alpine region of Australia for those who like bush walking (Kosciuszko National Park). 

IMAGE: Photomontage of rendering of next step fusion experiment ITER, taken from Australian ITER forum (https://fusion.ainse.edu.au/)

Although these skills can be acquired over the course of the project, familiarity with compilation of Fortran and/or C++ via LLVM will be a plus. Familiarity with bounded model checking (e.g., using CBMC) also a bonus, and otherwise something to look forward to learning. And of course, if you are already a plasma physicist, then that likely puts you ahead-of-the-curve when interpreting codes and their outputs.

Some supervisors:

- Jesse Cranney (Research School of Astronomy and Astrophysics),

- Charles Gretton (School of Computing)

The Giant Magellan Telescope (GMT) requires real-time phasing of its primary mirror segments. Each instrument of GMT is required to provide phasing telemetry from guide-star images. The required phasing data is embedded in these images in a highly non-linear way. Neural-network based solutions ought to be capable of extracting this data efficiently, and in a way that is tuned online to perform optimally under time-varying conditions.

Learning outcomes for a student undertaking this project include:

• Proficiency in real-time data processing and analysis.
• Skill development in utilising neural network-based solutions for efficient data extraction.
• Understanding of complex data structures and their analysis.
• Application of adaptive techniques for optimal performance in dynamic conditions.
• Competency in Python programming is required. Prior experience in control and estimation  theory, Machine Learning, and Fourier Optics is valued.

Background

We have been collaborating prototyping new network-based workflows for astronomy instrumentation. For some background reading, checkout our recent papers:

Smith et al., “Enhanced Adaptive Optics Control with Image to Image Translation”, UAI 2022.

Smith et al., “ Image-to-image translation for wavefront and point spread function estimation “, Journal of Astronomical Telescopes, Instruments, and Systems, Vol. 9, Issue 1, 019001 (January 2023). https://doi.org/10.1117/1.JATIS.9.1.019001

IMAGE: Is from https://mavis-ao.org/mavis/

Some experience training and evaluating an RL agent would put in you a good position.

No astronomy background required, although of course that will give you a special appreciation for what we are doing.

Some years ago M, Anjos [*A] characterised and evaluated semi-definite programming (SDP) relaxations of the Boolean SAT problem. His empirical work was interesting and somewhat promising, but did not obviously impact the course of fashionable decision procedures for the SAT problem. There are related works that consider optimisation problems that can be posed about formulae in Boolean logic [*B]. Since Anjos’ study early this century there have been enormous advances in heterogeneous computing, and particular GPGPU libraries and hardware. On that basis, we expect revising the SDP characterisation of SAT will be timely, and yield some interesting results compared to serial and incremental systematic searches that make use of systematic lookahead and CDCL-style backtracking search procedures.

Goals

This project suggests exploring the use of existing implementations of an SDP solution procedure, and swap out the BLAS and LAPACK libraries with their CUDA/GPGPU counterparts. For example, using “libnvblas.so” for the BLAS implementation. The target  problem corpus will be “hard” random UNSAT problems in the 3-SAT family, with upwards of 700 Boolean variables.

Background literature

[*A] Anjos, M. An improved semidefinite programming relaxation for the satisfiability problem. Math. Program. 102, 589–608 (2005).

[*B] Carla P. Gomes, Willem-Jan van Hoeve, and Lucian Leahu. The Power of Semidefinite Programming Relaxations for MAX-SAT. In Proc. Conference: Integration of AI and OR Techniques in Constraint Programming for Combinatorial Optimization Problems, Third International Conference, CPAIOR 2006, Cork, Ireland, May 31 - June 2, 2006.

[*C] Helmberg C., Rendl F., Vanderbei R.J., Wolkowicz H.: An interior-point method for semidefinite programming. SIAM J. Optim. 6, 342–361 (1996)

[#A] https://github.com/coin-or/Csdp - A C implementation of the predictor-corrector version of the primal-dual barrier method of Helmberg at al. [*C].

Confident in programming and logic, and an interest in empirical computer science.

Efficient CEGAR-tableaux for non-classical logics

Classical propositional logic (CPL) captures our basic understanding of the linguistic connectives "and", "or" and "not". It also provides a very good basis for digital circuits. But it does not account for more sophisticated linguistic notions such as "always", "possibly", "believed" or "knows".  Philosophers therefore invented many different non-classical logics which extend CPL with further operators for these notions.

Over the past fifty years, such non-classical logics have proved vital in computer science and logic-based artificial intelligence: after all, any intelligent agent must be able to reason with more than just and-gates, not-gates and or-gates! For example, we now have non-classical logics which capture notions such as "phi is true until psi becomes true"; "if we execute atomic program alpha in state x then we end up in a state y which obeys formula psi", and many more.

Of course, there is no such thing as a free lunch, for it is significantly harder to decide whether a non-classical formula is true or false (EXPTIME vs NP). Can we develop and implement efficient algorithms for this problem?

This problem has been attacked using multiple different methods for the past 40 years, without much success. But, in 2021, second-year student Cormac Kikkert and I showed that we could improve the existing decision procedures for these logics by orders of magnitude by combining two well-known existing methods giving rise to what we called CEGAR-tableaux [1].

Your project is to continue this work and to hopefully publish an academic paper in an international conference in 2024.

You will need excellent programming skills and a good background in maths. This project would set you up for a follow-up honours project in this area, which is exactly what Cormac is currently doing!

https://github.com/cormackikkert/CEGARBox

https://github.com/cormackikkert/CEGARBoxCPP

Rajeev Goré, Cormac Kikkert: CEGAR-Tableaux: Improved Modal Satisfiability via Modal Clause-Learning and SAT. TABLEAUX 2021: 74-91.

Rajeev Gore and Cormac Kikkert cegarboxktn : Improved decision procedures for multimodal tense description logic ALCI. (submitted to TABLEAUX 2023).

Rajeev Gore, Cormac Kikkert: Improved modal satisfiability for KD, KT, KB, K4 and K5 via CEGAR-tableaux: system description (submitted to TABLEAUX 2023).

• Excellent programming skills
• Strong maths background
• Some previous exposure to logic courses
• Enthusiasm!

The shortest-path distance query problem is fundamental problem in graph theory and has numerous applications in domains such as World Wide Web. Concretely, given any two vertices in a graph, the shortest-path distance query problem finds the shortest-path distance between these two vertices in the graph. The k shortest-path distance query problem is a generalization of the classical shortest-path distance query problem in a given network. It finds not only one shortest-path distance but also finds next k − 1 shortest-path distances. Not all k path distances returned are necessarily shortest-path distances.

Finding k shortest paths is possible by extending Dijkstra’s algorithm or Bellman-Ford algorithm. However, they may take several seconds to answer just a single k shortest-path distance query. Thus, they are not suitable to apply in real-time applications which require results to be provided in the order of microseconds. In this work, we aim to design scalable and efficient algorithms to answer k shortest-path distance queries which can promise a good trade-off between query time, indexing time and indexing size. During this project, we will be working on exciting ideas and exploring the searching and indexing techniques.

The ideal candidate should have

• a very good background in algorithms and theoretical computer science
• excellent programming skills in C/C++

Building secure electronic voting systems is hard and necessary. Trust is elections is essential to democratic society and currently at dangerously low levels in parts of the western world.

The Trust The Vote Project by the Open Source Election Technology Institute aims to address this by building a new electronic voting system; given the past record of new electronic voting systems it is likely that this system contains serious flaws.

This project involves examining the system to look for common flaws.

This project should be suitable for any computing student. Ideal background includes:

• knowledge of python
• some knowledge of cryptography.

Distributed acoustic sensing (DAS) is a new technique for monitoring and detecting geological processes such as earthquakes. A specialized photonics system repurposes buried fibre optic cables to detect vibrations in the Earth at unprecedented scale. DAS observations produce a terabyte of data per day and require high performance computing (HPC) to store and process.  A current problem is that due to the high sensitivity of DAS, road traffic and other anthropogenic signals interfere with observations of small and weak earthquakes.

Recently, researchers have looked at using image segmentation techniques in order to tease apart anthropogenic signals from the observed DAS data to improve the detection of small earthquakes. This project would look at the application of Curvelet Transform to DAS observations from Melbourne (city) and New Zealand (rural). This project would involve HPC (at NCI) and potentially GPU programming to ensure that data can be processed in an efficient manner.

This is a joint project with the Research School of Earth Sciences and will be co-supervised by Dr Voon Hui Lai.

At a minimum, experience in Python/Matlab/C/C++ programming.

Mathematical/Physics/HPC experience is desirable.

Video captures motions such as natural dynamics and human actions. Various research works have been dedicated in learning and extracting spatio-temporal features for scene understanding, human action recognition, anomaly detection, etc. Nowadays, video data have been preprocessed using either machine learning tools / computer vision algorithms or physical sensors, such as optical flows that highlight the dynamics, depth videos that segment the foreground objects or even human subjects, and skeleton sequences that focus on human actions, for better video understanding. However, videos contain very redundant information which hinders the effectiveness and efficiency in extracting the motions, forcing the computer vision community works hard in various large-scale pre-training, and suffers the dark matter of large models for a long time. In this research project, we aim to develop more lightweight video data formats which are able to efficiently distill the motions at different granular levels by removing redundant information while focusing on the core dynamics. Many downstream video processing tasks will benefit from our video dynamic distillation process, towards making video understanding much easier and more efficient. This also opens up a new research direction in exploring better video data representations for more lightweight cutting-edge video models.

This project is to explore and develop efficient video representations that would benefit downstream tasks such as action recognition, anomaly detection, scene understanding, etc. Skills in Matlab and/or Python is beneficial. The project will use machine learning and computer vision techniques.

Graph algorithms provide fundamental and powerful ways of exploiting the structure of graphs. Recently, graph neural networks (GNNs) have also become popular models for solving graph related learning tasks. Various research attempts have been made to understand the connections between the representation power of GNNs and graph algorithms such as the Weisfeiler-Leman algorithm. Through these connections, a beautiful characterisation of GNNs by finite-variable logics has been established. Nonetheless, some interesting questions still remain under-explored, such as the locality of GNNs and first-order logic/finite model theory, the generalisation ability and the interpretability of GNNs, etc. In this project, one specific aspect relating to graph algorithms and graph neural networks will be explored.

The project will be hosted at Graph Research Lab@ANU: https://graphlabanu.github.io/website/

Background literature:

A New Perspective on “How Graph Neural Networks Go Beyond Weisfeiler-Lehman?”, A. Wijesinghe and Q. Wang, International Conference on Learning Representations (ICLR) 2022

Theory of Graph Neural Networks: Representation and Learning, Stefanie Jegelka, 2022

This project requires students to have a solid background in machine learning and algorithms (plus logic if students will be working on logic-related topics).

Network science is an interdisciplinary research field that employs diverse domains like graph theory, physics, sociology, and data mining to analyse complex systems. Complex systems are naturally modelled as graph-structured data, consisting of multiple components with non-trivial dependencies to each other. These systems can be found in many real-world applications, such as social media, epidemic networks, telecommunication systems, and ecosystems.  On the other hand, machine learning (ML) is focused on building algorithms and methods to make machines “learn” to do certain tasks such as similarity matching, data classification and clustering. While recent work has attempted to increase the generalisation ability and expressivity of ML models to a wide variety of tasks, the performance of them on the learning tasks of complex systems is not well explored. In this project, we will investigate the limitations of the existing ML models on tasks related to complex systems. Furthermore, we will develop more robust models that can capture nice properties and attributes of complex systems by utilising the concepts and theories of graph neural networks.

The project will be hosted by Graph Research Lab@ANU: https://graphlabanu.github.io/website/

Background literature:

S. Thurner, R. Hanel, and P. Klimek, Introduction to the theory of complex systems. Oxford University Press, 2018.

S. Ha and H. Jeong, “Unraveling hidden interactions in complex systems with deep learning,” Scientific reports, vol. 11, no. 1, pp. 1–13, 2021.

This project requires students to have good background knowledge in machine learning/deep learning and excellent programming skills.

The project studies a graph based diffusion model which mimics the propagation of (mis-)information on online social networks. The objective is to devise algorithms, which maximise the spread of a piece of information (motivated by viral marketing) or minimize the extent that some misinformation propagates.

1. Study various properties of the model, such as graph parameters which govern the spread.

2. Analyse the computational complexity of the aforementioned optimization problems.

3. Devise (approximation/randomised) algorithms for these problems.

4. Complement the theoretical findings by conducting experiments on real-world social networks data, such as Facebook and Twitter.

Please see the following paper as an example of research in this area:

Maximizing the spread of influence through a social network, D. Kempe, J. Kleinberg, and E. Tardos, KDD, 2003.

• A solid understanding of algorithm design techniques (such as approximation, randomized, and greedy algorithms). Familiarity with some ML algorithms is beneficial too.

• A good knowledge of complexity classes such as NP-hard.

• Familiarity with graph theory concepts (such as random graph models), and probability theory concepts (such as Markov chains).

• Excellent programming skills.

In particular, if you have received excellent marks in COMP3600, COMP4600, and MATH2301, you are a suitable candidate.

Large vision-language models such as CLIP and BLIP have been mind blowing in the computer vision and wider AI communities. While they have impressive performance and capabilities, research are mostly limited to their training schemes and model architectures. In this project, we will study problems associated with data, to fill in the knowledge gap facing the community. An example is how to optimize prompts to better use conditioned images. Another example is analyze the ambiguity residing in the resulting model output. Besides, it is also possible to study how to properly evaluate the quality of generated image captions and find ways of improving them. This project will contribute to the data-centric research activities in our group, where collaborations with PhD students and postdocs are highly encouraged. Ideally, outcomes of this project will be part of some future research submissions.

Students should have very strong programming and mathematics skills, and excellent communication and teamwork skills. Proficiency in pytorch is needed.

Figure out how to write the highest possible quality software! The CakeML project includes a verified compiler for a Haskell-like language, PureCake. Because the compiler is verified, programs you compile behave exactly as the language specification says they should. But: how do you know if what you've typed is what you *really* meant? To get around this, you need to be able to reason about the behaviour of user-programs. The project will involve figuring out a general methodology for doing this and/or some specific case-studies that will feed into the PureCake programmers' library.

Some programming experience in a functional language (Haskell, SML, OCaml, Scala); interest in logic, formal methods and interactive theorem-proving (e.g., Coq, Lean, Isabelle/HOL).

The project aims to revolutionise problem-solving in high-dimensional domains. We propose the integration of alternating direction methods with physics-informed neural networks (PINNs) to tackle complex, multi-dimensional problems modeled as differential equations.

By leveraging the power of PINNs, which combine deep learning with physical laws, we can effectively capture the underlying physics and constraints of the problem domain. The alternating direction approach further enhances the accuracy and efficiency by iteratively solving subproblems along different directions.

This novel methodology holds tremendous potential for a wide range of applications, including fluid dynamics, climate science, and finance, where high-dimensional problems are prevalent.

Our project will involve developing and training specialized neural network architectures, designing efficient direction splitting algorithms, and conducting experiments to validate the effectiveness of our proposed approach.

Ultimately, our goal is to provide a robust and scalable solution for solving high-dimensional differential problems, enabling advancements in various scientific and engineering fields. This project represents a significant step towards unlocking new possibilities and pushing the boundaries of problem-solving capabilities.

S1. Programming experience in Python, Julia, Matlab, C++, or Fortran.

2. Basic Calculus and linear algebras, including derivatives, integrations, and matrix operators.

NoSQL databases (e.g., Key-Value or KV stores) are the backbone of modern data-intensive services at corporations, such as, Meta, Google, Twitter, and Linkedin. A critical data structure that KV storage engines use is the log-structured merge (LSM) tree. LSM trees are especially used for write-intensive workloads as insertions in LSM happen in an in-DRAM buffer. These DRAM buffers are sorted and written (flushed) sequentially to disks where they are compacted via merge sort to remove duplicates and for storage efficiency.

Today, compaction costs CPU cycles and is memory and I/O intensive. In this project, we will fundamentally rethink compaction strategies by deeply understanding their microarchitectural behavior. We will gain an understanding how compaction strategies interact with the features of a modern out-of-order processor. The outcomes will inform the design and implementation of more efficient compaction strategies.

We will use Meta's RocksDB for our study. The student should be comfortable (or be able to quickly learn) C++ programming.

When machine learning is applied to materials science complex physical and chemical structures are represented by structural features rather than positions of atoms and bonds. Similarly, when modelling such complex physical systems levels of detail can be removed to produce coarse-grained models that are more computationally viable.

Both these transformations are one-way mappings that lose degrees of freedom (i.e. information) but preserve specific key information. While the reverse mapping is not unique, there are high impact applications in which generating a reverse coarse grained model is desirable.

These include stream active learning to direct experiments, hierarchical models, and materials discovery. This project will investigate inverse and generative models to apply reverse coarse graining and feature-to-coordinate transforms to nanostructured materials data.

This project will focus on adapting and implementing algorithms in Python. The student will be provided with nanomaterial datasets of varying dimensionality and complexity.

In undertaking this project students will be required to build an understanding of the assumptions and limits of the implemented methods. The student will be required to use Python and document results and workflows using Jupyter notebooks, a written report and a GitLab repository.

Prior experience in Python and with machine learning methods is ideal but students with analytic skills and coding experience who are interested in gaining this experience are encouraged.

Prior experience with deep learning, generative models or atomistic material structures is not required but there is potential for student experience or interest to direct the project focus.

The current probabilistic federated learning benchmarks are toy tasks with little relevance to real-world problems. This project will develop new benchmarks with an emphasis on the ability to support real-world decision-making and analyse the quality of the predictive uncertainty for different federated learning methods, and the impact of approximations on downstream decisions such as identifying difficult examples. We will also significantly improve the existing benchmarks to consider federated training of deep networks using a large number of distributed data sources with an inhomogeneous distribution of data points, features and labels.

A solid machine learning and statistics foundation would be ideal.

Self-attention has become a common building block in many deep learning architectures but requires high memory and compute. This project will investigate several state-of-the-art efficient attention implementations, their approximation quality, and their impact on down-stream performance.

A solid machine learning and statistics foundation would be ideal.

Artificial photosynthesis of hydrocarbons from carbon dioxide (CO2), using water as the reducing agent and sunlight to drive the reaction, has the potential to provide 'renewable' fuels and chemicals at scale, replacing fossil fuels in a range of industries.

This project will explore new ways to produce important chemical products from sunlight, using the unique optical properties of nanoscale metal structures.

So called 'plasmonc' metal structures can be designed to interact strongly with light, leading to optical resonances at well defined energies that confine light in very small volumes. We leverage these so called plasmonic resonances to control light driven chemical reactions. Metal nanostructures can act as both light absorbers and catalysts and can enhance reaction rates at low operating pressures and temperatures. In addition, optically exciting molecules that are strongly bound to metal nanostructures can activate new reaction pathways, providing the opportunity to drive reactions at lower energies, or favour one reaction over another.

The project can be tailored to students interests and could include experimental work, computer simulations, and/or analytical modelling. An interest and aptitude in at least one of the following is required: chemistry or chemical engineering, material science, physics, optics,and energy.

Two-dimensional (2D) nano-materials, such as molybdenum disulfide (MoS2) and graphene, have atomic or molecular thickness, exhibiting promising applications in nano-electro-mechanical systems. Graphene is a one-atom thick carbon sheet, with atoms arranged in a regular hexagonal pattern. Molybdenum disulfide (MoS2) belongs to transition metal dichalcogenides (TMD) semiconductor family YX2 (Y=Mo, W; X=S, Se, Te), with a layered structure. This project aims to demonstrate novel opto-electronic devices, like light-emitting diode (LED), solar cells, etc. These 2D nano-materials can also be integrated into nano-electro-mechanical systems, enabling ultra-sensitive mechanical mass sensors, with single molecule or even single atom sensitivities. Moreover, the mechanical resonators based on these 2D nano-materials would be a perfect platform to investigate quantum mechanics, opto-mechanics, material internal friction force, nonlinear physics, etc.

Requirements/prerequisites:

• Students should be passionate about experiments and nanotechnology
• Priority is given to students who have a minimum GPA of 6.0/7.0 (or very similar)

Wearable devices are shaping this digital world. It will endow the world's highest intelligent creatures -humans with new attributes such as digitization, Internet of Things (IoT), quantitative sensing and detection. Green and sustainable energy supply for flexible wearable devices is a challenging and crucial research frontier. This project will focus on the cutting-edge power generation approaches, behind which physics of various energy conversion mechanisms. Based on application scenarios in healthcare, industrial inspection, structural monitoring, armed forces and consumer electronics, etc., a system architecture of the wearable flexible system is to be designed and tested. It is expected to make breakthroughs and reshape the digital world by developing all-in-one printable wearable electronics, self-powered self-aware wearable system, hybrid-integrated system on chip for flexible electronics, and IoT-enabled self-contained system towards full life cycle monitoring.

• Students should be passionate about experiments and nanotechnology
• Priority is given to students who have a minimum GPA of 6.0/7.0 (or very similar)

Single photon sources are critical for future quantum technologies such as quantum computing, quantum simulators, and unconditionally secure quantum communication.

The recent discovery of quantum emitters in two-dimensional (2D) materials offers a very promising source of single photon sources, with compelling applications for the next generation of integrated photonic devices. In contrast to their 3D counterparts, quantum emitters in amotically thin 2D lattices are not surrounded by any high refractive index medium. This eliminates total internal and Fresnel reflection of emitted single-photons, allowing intrinsically near-ideal extraction efficiency.

Quantum emission has been reported from a diversity of materials, in semiconducting transition metal dichalcogenides (TMDs) and insulating hexagonal boron nitride (hBN). The large band gap of the latter even allows one to resolve the zero phonon line (ZPL) at room temperature and thwarts non-radiative recombination of the localized exciton. Thus, single-photon emitters in hBN have an intrinsically high quantum efficiency which leads to significantly brighter emission. These single-photon sources are suitable for many practical field applications due to their resistance to ionizing radiation, temperature stability over a range spanning 800 K, long-term operation and capabilities for integration with photonic networks, as well as easy handling.

[1] ACS Photonics 6, 1955 (2019)

[2] Nanoscale 11, 14362 (2019)

[3] Nat. Commun. 10, 1202 (2019)

[4] ACS Photonics 5, 2305 (2018)

[5] J. Phys. D 50, 295101 (2017)

Requirements/prerequisites:

• Students should be passionate about experiments, nanotechnology and quantum science
• Priority is given to students who have a minimum GPA of 6.0/7.0 (or very similar)

The key challenge in realising an Australia-wide Electric Vehicle (EV) deployment lies in integrating the EV fast-charging infrastructure into power systems with resilience, safety, and efficiency guarantees. This research project aims to develop new control and optimisation algorithms that are necessary for the robust operation of electricity grids with integrated EV fast-charging infrastructure operated in coordination with wind and solar-generated electricity. The expected benefits of the research outcomes include enabling significant reduction in carbon emissions from the transportation sector, accelerating the energy transition to renewables, and placing Australian industry at the forefront of EV grid integration technology.

An interest in control and/or the electric power grid.

We live in a world where most of our actions are constantly captured by cameras. Video cameras are spread almost everywhere: in smartphones, computers, drones, surveillance systems, cars, robots, intercoms, etc. The quality of video data highly affects the performance of video models in its massive useful applications. Video data optimisation aims to improve the quality of video data being feed into the model for training hence improving the model performance, e.g., recognition accuracy and generalisation abilities. However, videos are more challenging due to the extra introduced temporal signals, which makes video understanding hard. Compared to image domain, videos suffer from much more serious data quality issues. Moving cameras, challenging natural dynamics e.g., rainy, snowy, reflection, etc., background dynamics e.g., tree waving, camera noise and camera shaking issues degrade the video model performance when trained on these unstable and noisy videos. Further challenging issues including the label errors and noisy ground truths, and the labeling process for video contents is even more labor-intensive and tedious. Many video data optimisation methods have been proposed in the literature for training robust video models, for example, video data augmentation is widely used in video model training. Generic video augmentation uses some basic video transformations such as geometric, color space, temporal, erasing and mixing, e.g., video data augmentation via temporal cropping. However, these augmentation methods are still unable to address the video data quality issues as they simply create a large number of diverse video contents with different focuses which still highly rely on the quality of original video data. In this research project, we aim to explore video data optimisation techniques that are able to improve the model performance for downstream video processing tasks such as human action recognition and anomaly detection. We also explore (i) how each individual training video affect the generalisation ability of a model and (ii) do we need all the video data in the training set or shall we drop unfavorable video samples and how.

Good programming experience in Python or Matlab is required. Experience in computer vision is desirable.

The unprecedented increase in wireless data traffic is advocating the investigation of suitable electromagnetic spectrum for the sixth generation (6G) and beyond wireless systems. The ultra-wide terahertz band ranging from 0.1 to 10 THz has recently attracted rapidly growing attention from academia and industry, due to its unique and huge potential to support ultra-fast terabit-per-second transmission. Built on the recent advancements in terahertz hardware and standardisation, the student in this project will focus on designing new communication strategies and signal processing algorithms for enabling intelligent THz communication systems for 6G and beyond wireless systems.

Solid background in communications and signal processing, high level programming skills (such as the ability to write numerical simulation programs in MATLAB or equivalent and/or machine learning programs in Python or equivalent), excellent written and verbal communication skills, and good interpersonal skills.

High-temperature thermal energy storage (TES) plays a crucial role in the development of cost-effective decarbonization technologies. It finds application in various processes such as heating hydrogen and/or iron ore in H2-DRI (400-900°C), iron ore beneficiation (400-800°C), alumina calcination (1000°C), and lime calcination (900-1000°C). However, there is a lack of detailed information regarding the thermo-physical properties, stability, and durability of potential storage materials at the relevant operating temperatures.
To address this issue, our team is engaged in identifying suitable storage materials by collecting data from vendors and the existing literature.

This project aims to measure the thermo-physical properties of materials through laboratory-based characterization techniques, specifically focusing on key materials with data gaps. Ultimately, this research will aid in advancing low-cost decarbonization technology pathways and facilitating the transition towards a sustainable and decarbonized future..

Students should have an:
- Interest in materials engineering and characterization;
- a desire to learn new experimental methods and
- good communication skills.