A5026408915- Domain Adaptation and Few-Shot Learning
- Single-cell and spatial transcriptomics
- Cell Image Analysis Techniques
- Solar Radiation and Photovoltaics
- COVID-19 diagnosis using AI
- Image Retrieval and Classification Techniques
- Radiomics and Machine Learning in Medical Imaging
- AI in cancer detection
- Energy Load and Power Forecasting
- Data Visualization and Analytics
- Balance, Gait, and Falls Prevention
- Multimodal Machine Learning Applications
- Gene expression and cancer classification
- Statistical and Computational Modeling
- Photovoltaic System Optimization Techniques
- Time Series Analysis and Forecasting
- Computational Drug Discovery Methods
- Lower Extremity Biomechanics and Pathologies
- Neural dynamics and brain function
- Plant Virus Research Studies
- Model Reduction and Neural Networks
- Inertial Sensor and Navigation
- Genetics, Bioinformatics, and Biomedical Research
- Nutrition and Health in Aging
- Smart Agriculture and AI
The University of Melbourne
2019-2025
University of Technology Sydney
2021
Regional solar power forecasting, which involves predicting the total generation from all rooftop photovoltaic (PV) systems in a region holds significant importance for various stakeholders energy sector to ensure stable electricity supply. However, vast amount of and weather time series geographically dispersed locations that need be considered forecasting process makes accurate regional challenging. Therefore, previous studies have limited focus either single (i.e., aggregated series) is...
Automatically converting text descriptions into images using transformer architectures has recently received considerable attention.Such advances have implications for many applied design disciplines across fashion, art, architecture, urban planning, landscape and the future tools available to such disciplines.However, a detailed analysis capturing capabilities of models, specifically with focus on built environment, not been performed date.In this work, we investigate biases text-to-image...
Catastrophic forgetting; the loss of old knowledge upon acquiring new knowledge, is a pitfall faced by deep neural networks in real-world applications. Many prevailing solutions to this problem rely on storing exemplars (previously encountered data), which may not be feasible applications with memory limitations or privacy constraints. Therefore, recent focus has been Non-Exemplar based Class Incremental Learning (NECIL) where model incrementally learns about classes without using any past...
Abstract Tumour-Infiltrating Lymphocytes (TILs) are pivotal in the immune response against cancer cells. Existing deep learning methods for TIL analysis whole-slide images (WSIs) demand extensive patch-level annotations, often requiring labour-intensive specialist input. To address this, we propose a framework named an notation-efficient s egmentation and ttention-based c lassifier (ANSAC). ANSAC requires only slide-level labels to classify WSIs as having high vs. low scores, with binary...
Non-parametric dimensionality reduction techniques, such as t-SNE and UMAP, are proficient in providing visualizations for datasets of fixed sizes. However, they cannot incrementally map insert new data points into an already provided visualization. We present Self-Organizing Nebulous Growths (SONG), a parametric nonlinear technique that supports incremental visualization, i.e., addition while preserving the structure existing In addition, SONG is capable handling increments, no matter...
Dual-energy X-ray absorptiometry (DXA) scans are one of the most frequently used imaging techniques for calculating bone mineral density, yet fracture risk using DXA image features is rarely performed. The objective this study was to combine deep neural networks, together with images and patient clinical information, evaluate in a cohort adults at least known fall age-matched healthy controls. entire body as, well as isolated hip, forearm, spine (1488 total), were obtained from 478 fallers...
Abstract Longitudinal studies that continuously generate data enable the capture of temporal variations in experimentally observed parameters, facilitating interpretation results a time-aware manner. We propose IL-VIS (incrementally learned visualizer), new machine learning pipeline incrementally learns and visualizes progression trajectory representing longitudinal changes studies. At each sampling time point an experiment, generates snapshot process on thus far, feature is beyond reach...
Neural growth is the process of growing a small neural network to large and has been utilized accelerate training deep networks. One crucial aspect determining optimal timing. However, few studies investigate this systematically. Our study reveals that inherently exhibits regularization effect, whose intensity influenced by chosen policy for While effect may mitigate overfitting risk model, it lead notable accuracy drop when model underfits. Yet, current approaches have not addressed issue...
Contemporary single-cell technologies produce data with a vast number of variables at rapid pace, making large volumes high-dimensional available. The exploratory analysis such high dimensional can be aided by intuitive low visualizations. In this work, we investigate how both discrete and continuous structures in single cell captured using the recently proposed dimensionality reduction method SONG, compare results commonly used methods UMAP PHATE. Using simulated real-world datasets,...
Regional solar power forecasting, which involves predicting the total generation from all rooftop photovoltaic systems in a region holds significant importance for various stakeholders energy sector. However, vast amount of and weather time series geographically dispersed locations that need to be considered forecasting process makes accurate regional challenging. Therefore, previous work has limited focus either single (i.e., aggregated series) is addition region, disregarding...
Neural growth is the process of growing a small neural network to large and has been utilized accelerate training deep networks. One crucial aspect determining optimal timing. However, few studies investigate this systematically. Our study reveals that inherently exhibits regularization effect, whose intensity influenced by chosen policy for While effect may mitigate overfitting risk model, it lead notable accuracy drop when model underfits. Yet, current approaches have not addressed issue...
Traditional machine learning is generally treated as a black-box optimization problem and does not typically produce interpretable functions that connect inputs outputs. However, the ability to discover such desirable. In this work, we propose GINN-LP, an neural network form coefficients of underlying equation dataset, when assumed take multivariate Laurent Polynomial. This facilitated by new type block, named “power-term approximator block”, consisting logarithmic exponential activation...
Estimating the parameters that describe ecology of viruses,particularly those are novel, can be made possible using metagenomic approaches. However, best-performing existing methods require databases to first estimate an average genome length a viral community before being able other parameters, such as richness. Although this approach has been widely used, it adversely skew results since majority viruses yet catalogued in databases.In paper, we present ENVirT, method for estimating richness...
Abstract Tumor-Infiltrating Lymphocytes (TILs) play a crucial role in the immune response against cancer cells. The existing deep learning methods for TIL analysis whole-slide images (WSIs) require extensive annotations at sub-image (patch) level, which often time-consuming and labor-intensive specialist input. To address this, we propose first-of-its-kind framework named annotation-efficient segmentation attention-based classifier (ANSAC). ANSAC requires only slide-level labels to classify...
Catastrophic forgetting; the loss of old knowledge upon acquiring new knowledge, is a pitfall faced by deep neural networks in real-world applications. Many prevailing solutions to this problem rely on storing exemplars (previously encountered data), which may not be feasible applications with memory limitations or privacy constraints. Therefore, recent focus has been Non-Exemplar based Class Incremental Learning (NECIL) where model incrementally learns about classes without using any past...
Traditional machine learning is generally treated as a black-box optimization problem and does not typically produce interpretable functions that connect inputs outputs. However, the ability to discover such desirable. In this work, we propose GINN-LP, an neural network form coefficients of underlying equation dataset, when assumed take multivariate Laurent Polynomial. This facilitated by new type block, named "power-term approximator block", consisting logarithmic exponential activation...