Our methodological work reflects the fact that practitioners of dynamic modelling for public health currently face many hurdles and challenges in building, validating, executing, modifying and understanding their models, and face needless hurdles and barriers in combining such models with data science techniques and to leverage sources of health big data. Several lines of work I am presently pursuing reflect the heavy performance burden of simulating large-scale populations at the individual level, and the high complexity of understanding the behaviour of such models.
- Applying category theory in novel ways to dynamic modeling and system science more generally.
- Design of novel programming languages for individual-based dynamic modeling. We are particularly pursuing application of functional reactive programming, in ways that provide broad support for concise, modular, transparent, declarative and expressive models, and allow for easy combination with machine learning/computational statistical tools such as Particle Filtering/SMC and PMCMC.
- Use of GPU and distributed computing for particle filtering and particle MCMC with dynamic models.
- Development of scale-modelling techniques for speeding the simulation and analysis of individual-based models. This work involves the first application of principles of dimensional analysis to the design of simulation models. The approach has attracted considerable industry interest.
- Development of formal methods for analyzing the behaviour of individual-based models. Much of this work is aimed both at facilitating understanding of individual-based models and for assisting with model order reduction -- trying to formulate simpler models that capture the same essential dynamics, but are computationally frugal and easier to manipulate, evolve and understand.
- Large-scale smartphone/mobile wireless sensor network-based data collection for informing the parameterization and calibration of individual-based simulation models. Our iEpi and Ethica epidemiological smartphone and wearable-based mobile data collection systems have emerged from this work. Much of this work is geared towards providing effective support for individual-based simulation models, seek to represent individual behaviour and interaction on which there is often limited data, and where self-report is unreliable. Wireless sensor mechaisms can offer benefits not only to researchers (making possible collection of far larger amounts of information for much lower resource investment), but also to policymakers for real-time epidemiological "signal detection".
- Application of other novel data collection techniques to inform dynamic understanding of health hehaviours, status and systems.We are particularly active in harvesting search, social media and web-based data that provide fine-grained temporal data across a wide range of channels.
- Creating tools to better manage and facilitate the dynamic modeling process.
- Exploring the characteristics of and tradeoffs between aggregate and individual-based models. This work helps modelling practitioners to better understand circumstances in which each type of modelling is desirable or required.
- Self correcting modelsLeveraging aspects of sequential Monte Carlo methods and PMCMC to create models whose estimates are regularly re-grounded in empirical data.
- Development and software implementation of mathematical tools for understanding the dynamic behaviors of simulation models.
- Creation of immersive software approaches for visualization and recognition of latent patterns in larger public health data sets, such as those based on virtual-reality systems.