Major Research Areas
The Evolutionary Computation and Machine Learning Research Group is active in many areas in evolutionary computation, machine learning and optimisation. In particular, we focus specifically on the following approaches:
Evolutionary Computation and Multi-objective Optimisation
This is the largest AI and machine learning research group in the southern hemisphere with a focus on evolutionary learning and optimisation. The research consists of evolutionary algorithms such as genetic algorithms (GAs) and genetic programming (GP), swarm intelligence such as Particle swarm optimisation (PSO), ant colony optimisation (ACO) and differential evolution (DE), and other population-based algorithms such as multi-objective optimisation (EMO) and estimation of distributed algorithms (EDA). We work in both evolutionary learning and evolutionary optimisation, particularly on discrete and combinatorial optimisation which has more applications to the real-world.
Deep Learning and Transfer Learning
We work on a number of different deep learning paradigms, including neural network based deep learning such as deep convolutional neural networks, deep auto-encoders, deep generative adversarial networks and deep belief networks and their variants, and non-neural network based deep learning such as GP-based deep learning, deep PCAs, and deep forest learning. This Group is particularly good in automated design of deep learning models, including deep neural networks and other deep models, where the structure/architecture and parameters are learned and optimised automatically and simultaneously.
Transfer learning is also a big area of this group, including domain adaptation, domain generalisation, and multi-task learning and optimisation. The work varies with different learning paradigms such as evolutionary computation techniques, neural networks, kernal based learning or symbolic learning algorithms.
Genetic programming (GP) is an important paradigm in evolutionary computation. Due to its flexible and powerful representation, it has been extensively used in symbolic regression, classification, clustering, feature selection and construction, computer vision and image analysis, AutoML, deep learning, transfer learning, scheduling and combinatorial optimisation. This group has been developing new GP methods and algorithms for for solving problems in all the above areas.
Intelligent Modelling and Symbolic Regression
Intelligent modelling and symbolic regression is a strategic direction of this group, aiming to discover mathematical, statistical and logic relationship between output variables (features/attributes) and input variables for a particular problem. Compared with conventional regression techniques where the data distributions need to be assumed/known, our techniques particularly genetic programming can automatically learn regression models without any assumption of data distributed, and often finding models that are interpretable to humans. This has been important for solving real-world application problems particularly in primary industry and (bio)medical and health tasks.
Computer Vision, Image Analysis and Pattern Recognition
Computer vision, image analysis and pattern recognition is a big area that this group has been focused on. Research in this big area has been involve in automatic image feature extraction, selection and construction, image and object classification, detection and tracking; edge detection, image segmentation, image understanding, and pattern recognition (classification, regression and clustering). This area has many real-world applications with huge economic and social value. Many funded research grants need research in this area.
Text Mining and Natural Language Processing
Text mining and natural language processing is a big area of AI. This group has been carrying out research in text minging, opinion mining, web mining, web intelligence, intelligent agents, natural language processing in medical data, hate speech detection, even detection, fake news detection, multi-view text classification. We have developing new deep neural networks, evolutionary computation algorithms, and other learning algorithms such as kernel-based methods to tackle these problems.
Scheduling and Combinatorial Optimisation
Scheduling and combinatorial optimisation is an important area in modern and high-value manufacturing, logistic and supply chain. We have been carrying research in job shop scheduling, vehicle and arc routing, web service composition, resource allocation, and timetabling. This group is particularly interested in hyper-heuristic approaches and solving dynamic problems under uncertain environment --- for solving problems in waste collection, mobile manufacturing, transportation, hospital timetabling, resource allocation and scheduling for cloud computing, logistic and supply chain scheduling.
Feature Selection, Construction and Dimensionality Reduction
In today's big data era, many applications involve data sets with thousands or millions of features/variables/attributes, and even with a small number of instances. This is a well known hard problem to solve in both AIML and statistics. In most scenarios, a big majority of these features are redundant or even useless. Feature selection aims to remove those useless and redundant features and select only a small number of important features (and accordingly significantly reduce the dimensionality of the data) that will not reduce and even improve the accuracy performance. Feature construction aims to construct/build a small number of new, high-level features (and accordingly further reduce the dimensionality) by combining those selected low-level features. Feature selection/construction have been applied to almost all pattern recognition/ML tasks and real-world applications.
Expainable/Intepretable AI and Machine Learning
Deep neural networks particularly deep convolutional neural networks and variations have achieved great results in image classification, speech recognition/signal processing, and natural language processing as well as some other domains. However, the (learned) models are too big and complex, and very hard to explain/interpret. In some applications where accuracy is the primary focus, this is perfectly fine; but for some other domains such as medical and legal areas where on top of accuracy why and how the learned models derive the solutions that can achieve good performance more more important than or at least as important as the accuracy performance itself. In such areas, learning good models that can be interpreted and understandable is important. The Group have been carrying out research in explainable AI (XAI) and machine learning that can derive interpretable models. The main approaches include feature selection/construction, online learning simplified models, designing interpretable components to learn interpretable models, post-hoc processing of learned complex models.
Application Areas of Evolutionary Computation and Machine Learning
This is a very big area of applications, particularly in NZ and many Asia-Pacific countries. This includes but not limited to aquaculture and open ocean for blue economy, agricultural planting and products, horticultural product, fruits and vegetables, plant/animal disease and weed, food safety and water resources. AI/ML techniques have started playing an important role in this direction, and a lot more work needs to be done.
Climate Change and Environment
This is critical for NZ and all the countries in the world. The main areas that need AI/ML techniques to make improvement include climate modelling, weather forecasting for land and ocean, earthquake prediction, volcano and tsunami prediction. water surface level prediction. This can also greatly help humans to go to dangerous scenarios.
Health, Wellbing, (Bio-) Medical Outcomes
This area of research is essential to human lives, and AI/ML can substantially improve performance and help domain experts to judge and evaluate diseases. This include medical diagnosis, cancer detection, biomarker detection, public health, healthcare systems, community diseases (e.g. SARS, COVID), drug discovery, etc. This is one of the most "popular" application areas in AIML.
High-tech Application Areas and High-value Manufacturing
This has many important applications in the high-tech area and high-value manufacturing. Examples include transportation, cyber-security, sustainable/renewable energy, electrical planning and scheduling, new materials (nano, elec device), superconductivity, mechanical and electrical manufacturing, block chain, IoT, Robotics, search based software engineering.
Economy, Social impact, Ethics, and Public Policies
This area appears in our daily lives. On one hand, the aspects on economic and financial modelling, banking, insurance, human resources and management, and tourism can all use AI/ML techniques to improve effectiveness and efficiency. On the other hand, in the ethical aspects and public policies AI and ML techniques should be used with extra care to avoid being misused to hurt humans and our communities. We have been conducting research in both aspects.
We also investigate the research platforms required for Evolutionary Computation research. We use a network of computers (Grid Computing) to develop a large number of our programs. This increases the range of tasks we can explore as well as improving the confidence of our results in any given domain. The potential of multiprocessor systems, e.g. general-purpose graphical processing units (GPGPUs) that are found on high-end computer graphics cards, is being explored.
For more information on our research, visit the publications page