I am interested to answer the question of how artificial intelligence and machine learning advances humanity, fuels the economy and sustains the ecosystem in the current digital transformation era.
- Machine Learning for Sensor Signal, Image and Information Processing
- Mathematical Theory and Algorithm Development of Machine Learning
- D-I-K-I (Data mining, Information Analytics, Knowledge discovery, Intelligence integration)
The core values of my research are discovery and impact. My vision is to develop new theories and demonstrable algorithms that behave with intelligence and to enable user system to yield optimum performance which is not possible without computational intelligence signal processing.
The focus is on the mathematical foundation of machine learning and artificial intelligence computing algorithms. Many signal and image processing algorithms have incorporated some forms of computational intelligence as part of its core framework in problem solving. These algorithms have the capacity to generalize and discover knowledge for themselves and learning to learn new information whenever unseen data is captured. Statistics is organised information but intelligence is more than that. We believe that it is just not enough to solve a problem by giving the most “frequent” answer. The goal is to develop new theories and demonstrable algorithms for deep learning machines that truly integrate signal processing and computational intelligence. These algorithms will have the ability to discover knowledge for themselves and learning to learn new information whenever unseen data is captured. While many researchers discount the need for human intervention in data analysis, our work reserves the freedom to enable collaborative syncretisation between human intuition and machine intelligence.
Research interest includes:
1. Machine Learning and Artificial Intelligence Computing
The focus is on the mathematical foundation of machine learning and artificial intelligence computing algorithms. Many signal processing and computer vision algorithms have incorporated some forms of computational intelligence as part of its core framework in problem solving. These algorithms have the capacity to generalize and discover knowledge for themselves and learning to learn new information whenever unseen data is captured. Statistics is organised information but intelligence is more than that. We believe that it is just not enough to solve a problem by giving the most “frequent” answer. The goal is to develop new theories and demonstrable algorithms for deep learning machines that truly integrate signal processing and computational intelligence. These algorithms will have the ability to discover knowledge for themselves and learning to learn new information whenever unseen data is captured. While many researchers discount the need for human intervention in data analysis, our work reserves the freedom to enable collaborative syncretisation between human intuition and machine intelligence. This includes:
- Exemplar-aided Machine Learning
- Hierarchical Deep Wide Artificial Neural Networks such as Convolutional LSTM Network
- Genetically-enabled Globally Optimized Ensemble Deep Learning Neural Networks
- Lifelong Learning Knowledge-based System
- DIKI-enabled Human-centric Self-insight AI
- Generative Adversarial and Collaborative Deep Nets
- Deep Reinforcement Learning, Multi-Agent Machine Learning
- Deep Neural Decision Trees
- Ethical AI
2. Data Mining, Signal Processing and Information Analytics
Data mining is the process of discovering patterns in large datasets involving methods at the intersection of machine learning, statistics, and database systems. It is an essential process where intelligent methods are applied to extract data patterns. Current research involves data pre-processing, model and inference considerations, interestingness metrics, complexity considerations, post-processing of discovered structures, visualization, and online updating. This includes:
- Latent Variable Analysis (LVA) which includes robust principal component analysis (PCA), independent component analysis (ICA), nonnegative matrix factorization (NMF), sparse coding (SC), compressive sensing (CS), low-rank sparse tensor decomposition, hybrid GMM-HMM models, time-varying switching models, nonlinear LVA models such as RBM, DBM and DBN
- Blind Signal Separation (BSS) where the mixing system is characterized as underdetermined, convolutive, noisy and possibly nonlinear. One of the earlier successes is the development of a mathematical framework based on Addition Theorem leading to a statistical multilayer neural network as the universal machine for solving the complex problem. Solutions to single-channel signal separation problems have also been developed by using hybrid LVA models with deep neural networks. Extraction of target signal buried in non-stationary narrowband background noise using exemplar-guided LVA models.
- Biometrics Pattern Recognition: It has been shown that biometrics data will vary depending on the how and when the data is captured and the environment it is subject to. By considering both pillars of nonlinearity and non-Gaussianity behaviours, research in biometrics has led to a new paradigm of thinking about pattern recognition problem. Multimodality biometrics include iris, sclera, face, palmprint, fingers texture, and physiological-based biometric such as heartwave captured at rest and during duress.
- Intelligent Computer Vision: Photometric stereo 3D depth estimation, image inference for object measurement in a photo, estimating depth of images based on single photo using deep neural network
3. Smart Sensing, Diagnostic and Data Monitoring
This theme focuses on creative and smart signal processing theories and methods to be applied for various engineering and IT applications including non-destructive testing (NDT), big-data analytics, wearable sensors, anomaly detection, and condition monitoring. The strategic trend in the development of smart sensing has changed towards the issue of safety in the broadest sense, to the protection of the population and the environment against man-made and natural disasters. There has been an increased realization of the potential benefits of applying advanced machine learning techniques to the signals captured from multimodal sensors. This includes:
- Intelligent Sensing & Computing for Non-Destructive Testing: Automatic defect detection, location, sizing & identification with ECPT using machine learning techniques with enhanced resolution (Robust PCA, ICA, NMF, Morphological Component Analysis, Low-Rank Sparse Decomposition, GA-enabled ICA-Images Fusion)
- Anomaly Detection using statistical process control techniques (PCA with T2 statistics, 2D contribution map, multiscale ICA-PCA for statistical process monitoring, multiple targets detection using hysteresis constant false alarm (h-cfa)), healthcare data analytics using deep learning and statistical process control
- Predictive Maintenance and Condition Monitoring: Sound event-based detection for railway fault, measuring and quantifying the impact of hard disk drive (HDD) noise on human, separating airborne & structural noises in ship structure for better noise-propagation reduction, reliability index estimation for ship control and operations
- Wearable Sensing: Wearable audio meter, continuous and non-invasive monitoring of anxiety and stress level in real-time situation, estimating autism quotient from empathizing quotient and systemizing quotient through game technology and artificial intelligence
4. Emerging Topics in United Nation Sustainable Development Goals
- Urban Energy Smart Grid for Affordable Clean Energy and Sustainable Cities: Energy flow is now very much a two-way process. Low carbon technologies generated by solar panels and wind farms are feeding into national power networks bringing new challenges to be overcome. Research is on-going to find a way of managing that power in real time, so that the low-carbon transition can be achieved at reasonable cost and without degrading power system reliability. The goal is to match supply to demand in real time and within network constraints, and that means making the grid more intelligent. This intelligence allows demand response, the involvement of customers, and energy storage to be integrated into existing networks.
Smart Energy Forecasting: Prediction of solar energy harvesting in PV and heat energy generation in TeG accounting for environment factors using artificial intelligence methods
Smart Energy Monitoring: Non-intrusive load monitoring using computational intelligence techniques, temperature-dependent lithium-ion battery model and state of charge estimation using Kalman filtering algorithms, Internet-of-Things (IoT) for home energy detection and home energy theft
Energy Storage Control: Fuzzy logic control of energy storage in grid-connected microgrid with renewable energies
- Precision Farming for Zero Hunger: Research in precision agriculture concentrates on farming management concept based on observing, measuring and responding to inter and intra-field variability in crops. Crop variability typically has both a spatial and temporal component which makes statistical/computational treatments quite involved. The holy grail of our research has been the ability to define a Decision Support System for whole farm management with the goal of optimizing returns on inputs while preserving resources. Presently, we are developing big data analytic tools to estimate the nutrient status of the plants to improve the efficiency of the fertilizer use.
- Bio-signal Analytics for Good Health and Well-Being: Designing new information analytic algorithms to tackle societal challenges related to health and ageing issues. Home monitoring makes use of bio-signals that are collected non-invasively, usually in self-help manner. These signals are not reliable and are subjected to motion artifacts, contact problems, poor quality sensors, shifted sensor locations, and signal interference from the environment. Advanced research is currently conducted to study the correlation of multi-modal bio-signals and to analyze them in combination so as to develop better methods for home monitoring. There are strong interests in developing context-based smart technologies where signals are processed in the local nodes of a body area network and then transmitted to the Internet (cloud) for predictive analytics.
Research Student Supervision Interests1. Machine Learning enhanced Nondestructive Anomaly Detection. 2. Deep Learning 3D Robotic Mapping. 3. Conversational AI Deep Chatbots. 4. AI-based Predictive Maintenance for Ion Beam Etching System. 5. AI-reinforced Single Channel Speech Mixture Separation. 6. AI-enabled wearable IoT for Long-Term Monitoring of Mental Health Wellbeing. 7. Self-Training AI for Autonomous Vehicles. 8. Ensemble AI for Intent and Emotion Recognition. 9. Automated real-time identification of cell structures in photomicrographs.
MSc, Communications Engineering
BEng (Hons), Electrical and Electronic Engineering