Research Interests

Keywords: human-centred AI systems, machine learning, agent-based systems, probabilistic graphical models, ontologies

My research explores the design of next generation human-centred AI systems. I take an augmented AI approach, where the human user works interactively and cooperatively with the AI system. In this scenario the system learns and adapts to the human user and in turn the user learns and adapts to the system. The primary design goal in augmented AI, is to amplify human cognitive power rather than to replace it.

Adaption and cognition are two integral aspects of augmented AI systems. Diverse AI techniques, such as machine learning, agent based systems, Bayesian networks and more broadly probalistic graphical models, and ontologies must be combined into hybrid AI systems that adapt to dynamic physical and social environments to support diverse human users with different application and decision-making contexts. I am especially interested in exploring new mechanisms for continual learning, and interactive scientific knowledge discovery and decision making in dynamic, complex (but bounded) physical or social environments. Examples of such systems include modeling and predicting the weather over some region, analysing patterns of household energy consumption behaviour in a country, monitoring and controlling indoor air quality, learning models for individualised and public health care, or understanding the dynamics of a stock market. Typically, sensors embedded in these systems continuously generate observational data which combined with expert domain knowledge allows us to gain some insight into the dynamics, i.e. the key processes and patterns that drive the system.

Within this context, I also have an interest in AI driven 3D digital twins and Internet of Things (IoT) systems. I feel that there is much work to be done at the intersection of AI systems and cyber-physical systems in general and that these communities are bound to converge. While, I have a strong interest social good applications in South Africa, I am also working with AI driven digital twins in the mining, vehicle manufacturing and finance sectors. I still maintain an interest in open systems and architectures for national health information systems in developing African countries, an area which I have worked in previously. You can read our new position paper on “Re-imagining health and well-being in low resource African settings” here. These projects are carried out within the Adaptive and Cognitive Systems Lab in the Centre for Artificial Intelligence Research (CAIR).

Current Projects

Topic 1 - Deep neural networks for spatial-temporal modelling

A number of deep learning architectures, such as Spatial Temporal Graph Neural Networks (ST-GNNs) have emerged recently for spatial-temporal (flow) modeling and prediction. These models, e.g. [1] are designed to model both temporal and spatial patterns and can be used for discovering insights into complex, dynamic and erratic systems. It can also serve as a powerful tool for automatic analysis of observations emanating from sensor networks deployed in such systems. Analysis tasks will include anomaly detection, data fusion and situation analysis. ST-GNNs are also able to capture and represent complex spatial-temporal dependencies from historical data observations. These techniques outperform other traditional DNN approaches such as the TCN and the BiLSTM. More details can be found in our recent papers where we applied ST-GNNS for share price prediction on the Johanesburg Stock Market (JSE) [2] and weather prediction in South Africa [3].

Topic 2 - Automatic machine learning

Bayesian optimisation has emerged as a leading technique for solving the combined algorithm and hyper parameter optimisation (CASH) problem [4]. However, other approaches that use traditional optimisation techniques from evolutionary computing methods and reinforcement learning for neural architecture search (NAS) have also shown promise in this area. This is a cross cutting topic and can either be a standalone topic or aspects of this can be introduced in topic 1, 3 and 4 or vice versa. We have a recent paper in this area [5].

Topic 3 – Scientific knowledge discovery and evolution

A key thrust of our current research is on designing novel AI systems that support scientific knowledge discovery and evolution (KDE). While routinely performed by human scientists, formalising and automating the KDE process is difficult. In Philosophy of Science (PoS), recent theories of method like the abductive theory of method (ATOM) [6] attempt to describe the process of discovering and justifying theories. ATOM encompasses a two-step process, i.e. phenomena detection from observations and then constructing and evaluating plausible theories to explain the detected phenomena. In our recent work we proposed a preliminary agent architecture for KDE based on ATOM. We have also published preliminary results on using the architecture to design a personal health care agent [7]. A key outstanding challenge and current focus area is theory construction and theory evaluation.

Topic 4 - Adaptive and cognitive agents and interactive decision making

While significant progress has been made in different branches of AI, e.g. machine learning, the design of real world systems that incorporate different AI techniques to deliberate about and adapt to erratic and changing environments remains an open challenge. Developing and deploying ML systems is relatively fast and cheap, but maintaining them over time is difficult and expensive [8]. From a software engineering perspective, the intelligent agent paradigm and agent oriented programming is a well established research area, but has not been widely adopted for designing and developing ML systems which are often developed in an adhoc piece meal fashion in practise. Architectures have also been proposed for intelligent monitoring systems e.g. [9, 10]. Next generation intelligent systems will combine techniques from different branches of AI, including intelligent agents, machine learning, Bayesian networks, ontology driven and rule based information systems. This topic will review and analyse research into agent oriented software engineering, intelligent systems, cogntive agents and architectures proposed in artificial general AI (AGI), towards developing reusable architectural patterns, frameworks and methodologies that can facilitate the design and development of knowledge discovery and decision making systems. See our recent papers on hybrid AI architectures for interactive decision making agents and systems for detecting heart disease [7], a cognitive decision making model for sugarcane growers [11] and investment decision maing on the JSE [12]. This is a cross cutting topic and can either be a standalone topic or aspects of this can be introduced in topics 1,2 and 3 or vice versa.

[1] J. Zhang, Y. Zheng, and D. Qi, “Deep spatio-temporal residual networks for citywide crowd flows prediction,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 31, no. 1, 2017, pp. 1655-1661. Link

[2] K. Pillay and D. Moodley, “Exploring graph neural networks for stock market prediction on the JSE,” Proceedings of the South African Conference on Artificial Intelligence Research (SACAIR), in Communications in Computer and Information Science, vol. 1551, 2022, pp. 95–110. Link

[3] M. Davidson and D. Moodley, “ST-GNNs for Weather Prediction in South Africa,” Proceedings of the South African Conference on Artificial Intelligence Research (SACAIR), in Communications in Computer and Information Science, vol. 1734, 2022, pp. 93–107. Link

[4] B. Shahriari, K. Swersky, Z. Wang, R. P. Adams and N. de Freitas, “Taking the Human Out of the Loop: A Review of Bayesian Optimization,” in Proceedings of the IEEE, vol. 104, no. 1, pp. 148-175, 2016. Link

[5] K. Kouassi and D. Moodley, “Automatic deep learning for trend prediction in time series data,” arXiv preprint, 2020. Link

[6] B. D. Haig, “An Abductive Theory of Scientific Method,” in Method Matters in Psychology: Essays in Applied Philosophy of Science, in Studies in Applied Philosophy, Epistemology and Rational Ethics, vol. 45, no. 4, pp. 35-64, Cham, Switzerland, Springer, 2005. Link

[7] T. Wanyana, M. Nzomo, C. S. Price and D. Moodley, “A Personal Health Agent for Decision Support in Arrhythmia Diagnosis,” revised selected paper, Proceedings of the International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2022), in Communications in Computer and Information Science, vol. 1856, 2023, pp. 385–407. Link

[8] D. Sculley et al., “Hidden Technical Debt in Machine Learning Systems”, Advances in Neural Information Processing Systems 28 (NeurIPS), 2015. Link

[9] J. Adeleke, D. Moodley, G. Rens, and A. Adewumi, “Integrating Statistical Machine Learning in a Semantic Sensor Web for Proactive Monitoring and Control,” in Sensors, vol. 17, no. 4, pp. 807, 2017. Link

[10] J. Adeleke, “A Semantic Sensor Web Framework for Proactive Environmental Monitoring and Control,” Ph.D. dissertation, School of Mathematics, Statistics and Computer Science, University of KwaZulu-Natal, Durban, South Africa, 2017. Link

[11] C. S. Price, D. Moodley, A. W. Pillay and G. B. Rens, “An adaptive, probabilistic, cognitive agent architecture for modelling sugarcane growers’ operational decision-making,” in South African Computer Journal, vol. 34, no. 1, pp. 152-191, 2022. Link

[12] R. Drake and D. Moodley, “INVEST: Ontology Driven Bayesian Networks for Investment Decision Making on the JSE,” presented at the South African Conference on Artificial Intelligence Research (SACAIR), Online, Dec. 6-10, 2021. Link

Scroll to Top