Statistical tools for Iron Ore Price. (MODERATOR Dr JANOSCH RIEGER)

In scenario generation of economic variables/indices, say the iron ore price, we can identify 15 other economic variables/indices as impacting on the iron ore price. For tractability of the forecast, the most important three variables should be selected as the input variables. Is there a generic ML method that can identify and remove the less impactful 12 variables as input variables? Put simply, how to distil from the 15 variables to only 3 variables that are then used to generate the future scenarios of the iron ore prices?

As a related question on scenario generation, we can identify some historical patterns (clusters) in a time series (of an economic variable), say 4 patterns are clearly identified and each pattern normally lasts for 3 years. The question is: after collecting data on the variable for 1 year, what kind of statistical method is more suitable for identify one out the 4 historical patterns as the most likely pattern reappearing now. For example, now we have got the time series data of the past 1 year, can we use a generic statistical method to analyse the 1 year’s data and to say the No.3 pattern is being repeated.

Patient Medication Adherence

The primary objective of this research project is to comprehensively review existing tools and programs designed to support patient medical adherence. (MODERATOR A/Prof TIM GARONI)

Subsequently, we aim to explore the feasibility of developing an innovative AI-driven tool that increases patient medical compliance and/or facilitates healthcare professionals in educating patients effectively. Patient medical adherence is a critical concern in healthcare, impacting the effectiveness of treatments and overall patient outcomes. Despite the availability of various tools and programs, non-adherence remains prevalent due to the inadequacy of existing solutions in addressing the diverse factors contributing to this problem. Inefficient communication channels between healthcare providers and patients, coupled with the lack of personalized and adaptive support systems, contribute to the persistence of non-adherence. Patients may struggle to consistently follow their prescribed regimens due to multifaceted challenges, including forgetfulness, lack of understanding about the importance of adherence, socioeconomic barriers, medication side effects, and the perceived burden of complex treatment plans. Additionally, issues related to inadequate communication between healthcare providers and patients, cultural beliefs, and mental health considerations further contribute to the pervasive problem of non-adherence. Recognizing and addressing these diverse factors is crucial in developing effective strategies and interventions to enhance patient adherence, ultimately improving health outcomes and the overall efficacy of medical treatments.
We need to understand more if this type of project is feasible. We propose to start with:
Conduct an extensive review of existing tools, applications, and programs designed to support patient medical adherence. Evaluate their effectiveness, limitations, and the underlying technologies.
Identify gaps in the current tools and programs that contribute to the persistence of non-adherence. Determine the key challenges faced by patients and healthcare professionals in achieving optimal medical compliance.

Explore the feasibility of developing an AI-driven tool to address the identified gaps. Assess the potential of artificial intelligence in personalizing adherence support, providing real-time feedback, and improving patient education.

There are two AI applications which we would like to explore:
True 3D segmentation of both anterior surfaces and retinal layers with the constraint that prediction times are of the order of no more than a few (< 5) seconds. (MODERATOR A/Prof. TIM GARONI)

Volumetric data is captured at a frame rate of 300 Hz, with each frame containing a sparsely sampled grid of 1008 depth profiles. Volumetric registration of sparsely sampled volumes poses a unique challenge. We would like to investigate alternative registration approaches, potentially including AI.
Cylite is a Melbourne based company developing and manufacturing leading edge 3D imaging technology for optometry and ophthalmology. Using highly parallelised optical coherence tomography, we image the eye with over 1000 simultaneous laser beams to produce true volumetric images free from patient movement artifacts.

What novel problems might be solvable by deploying a distilling step-by-step continuous learning model for multi-modality machine learning onto a modern mobile device? (MODERATOR: Dr IVAN GUO)

https://blog.research.google/2023/09/distilling-step-by-step-outperforming.html
https://blog.research.google/2023/09/on-device-content-distillation-with.html
https://blog.research.google/2023/06/speed-is-all-you-need-on-device.html
https://blog.research.google/2023/08/language-to-rewards-for-robotic-skill.html
Use anything at: https://blog.research.google/

Metro Trains Melbourne are required to meet Monthly Performance Targets as set out by the Government.
We would like to accurately forecast what our performance would likely be for the next 12 months based on our internal available data, external historical data such as weather, population growth etc. and future events? (MODERATOR: A/Prof TIM GARONI)

Metro Trains Melbourne moves passengers via rail across Melbourne. Our network includes 228 stations through inner city and greater Melbourne. Regional Rail Operator V/line and Freight trains use the Melbourne rail network to get through the city. Metro trains schedule trains including V/line and freight through our network. We have approximately 2500 services every day excluding V/line and freight services.

Metro Trains Melbourne’s are required to meet the Monthly Performance Targets as dictated by the Government:
Reliability: 98%, (No of trains delivered per day)
Punctuality: 92%, (No of trains ran on time up 04:59 late)
EOPR%: 97.37%, (measured based on punctuality and reliability and route length)

Metro captures the following data following an incident:
Date and time of incident
Location where the incident took place
Details of the incident including responsibility and cause of the incident
Punctuality and Reliability lost due to the incident
EOPR% lost due to the incident
Cancellations, trains bypassing the loop, trains terminated or originated before their planned destination, services affected, late services due to the incident

Causes that result in an incident which impact performance recorded internally which have been recorded for the past 5 years:
Police or PSO Operations
People entering the rail corridor for various reasons
People riding on the top or outside of the train
People behaving in an unruly or threatening manner at a station
Weather events
Special events
Number of Passengers at a station
Number of Assisted Boarding Passengers or Ill Passengers on the train
Train Faults such as faulty panto-graph, PA system faulty, etc.
Infrastructure faults such as broken rail, track circuit failure
Vandalism such as graffiti on train, objects thrown in pit
Resource limitations due to vacancies, absenteeism
V/line and Freight services travelling on the network

We also know the following future events with some certainty which impact our network:
Patronage levels
Weather (Seasonality)
Major Sporting events
Planned projects

Can a combination of mathematical modelling and machine learning enhance the signal to noise ratio of fluorescent images illuminated by an LED, to the levels obtained by laser illumination?
(MODERATOR: Prof. KAIS HAMZA)

Confocal Laser Endoscopy (CLE) is a technique used in medical imaging. A fluorescent dye is administered on the tissue being imaged. The sample is then illuminated by a laser. The sample reflects light at the same wavelength, while the dyes fluoresce at different wavelengths. Fluorescent and reflected light are captured by a photosensor. If the spectral width of the illumination source is sufficiently narrow, reflected light can be easily filtered out, leaving only the fluorescent light to form images.

The narrow spectral line-width is the primary reason why lasers are used in CLE. The second reason is the high intensity of laser light. However, lasers are bulky and expensive.

LEDs with high intensity have been developed in the recent past, but they are not as monochromatic as lasers. If used in CLE, the signal-to-noise ratio would be significantly poorer, leading to noisy images.

Optimal Energy Use for a Sail and Battery Powered Unmanned Ocean-Going Catamaran (Platypus Australia). Optimize the use of motors/energy Battery Powered Unmanned Ocean-Going Catamaran to perform a survey of an area. (MODERATOR Dr. VLADIMIRA SECKAROVA)

In this context, wind forecasts play a major role.
Optimization alternatives:
a) Earliest latest time to finish
b) Earliest expected time to finish
c) Other?

EV battery size optimization balancing critical minerals (this is a trade-off between the size of EV battery from a transport perspective and a use of EV in a power perspective which could make value for ovesizing) (MODERATOR Dr VLADIMIRA SECKAROVA)

TBA

Development of an optimization technique that works for a dual congestion policy for road transport and the power sector. (MODERATOR Dr VLADIMIRA SEKHAROVA)

TBA

Methodologies to directly or indirectly couple distribution networks into power systems models
and Optimization of transmission size and timing accounting for uncertainty in DER

TBA

Carrier / Sender location Optimization – which thereon builds inventory management (MODERATOR Dr JIE YEN FAN)

For a 4PL to identify the trends and optimise the sending location, carrier and inventory capacity based on modes of transport available, cost and transit.

Trends Identification for multi industry based customer group (MODERATOR Prof. KAIS HAMZA)

Methods or system for data to be manipulated and visualized for trends identification on peak, resource allocation etc along with better planning for carrier optimization.

Digital twin and the circular economy. (MODERATOR Dr CLAUDIA DEL CAMPO)

Developing a Circular Economy and Bioeconomy (CEBE) strategy is an action in the Government’s Emissions Reduction Plan (ERP) released in May 2022. Earlier this year the Ministry of Business,
Innovation and Employment (MBIE) released several Request for Proposal’s to develop New Zealand relevant knowledge about the Impacts, Barriers, and Enablers for a Circular Economy as an input to the development of the CEBE strategy. The research will help identify the most important or impactful ways in which Aotearoa New Zealand can shift to a more circular, low emissions economy, while taking account of our distinctive geographic, economic, ecological, and cultural context, managing the risk of negative outcomes, and realising opportunities that may arise. Detailed research on the bioeconomy (i.e., circular bioeconomy) is encompassed in this work. Project 4 of this investment in research is focused on enabling digital technologies for New
Zealand’s circular and bioeconomy, including the role of digital twins and is led by a consortium that includes the Sustainable Business Network, Aurecon, thinkstep-anz and Rākau Tautoko.
Specifically, the deliverables of Project 4 include –
• A high-level overview of developments in enabling digital technologies to manage data and information as it relates to the development of a circular economy.
• Assessing and demonstrating the strategic potential of digital twins for New Zealand