Research theme 03
How can machine learning, simulation, and scalable computation help us model and understand complex adaptive systems?
Agents feeding into a model representation and computational inference
Complex adaptive systems research runs into a practical wall quickly: interesting models are computationally expensive, and interesting questions about them — inference, calibration, rule discovery — multiply that cost. A recurring part of my work is closing that gap: taking a scientifically motivated model and making it fast and tractable enough to actually study.
That has meant re-implementing an agent-based simulation for CUDA to bring a multi-day run down to hours, and it now means exploring how machine learning — including evolutionary algorithms and transformer architectures — can accelerate model discovery itself. I am a co-author on a submitted DARPA MAGICS proposal exploring how transformer architectures combined with evolutionary algorithms could support generalizable evolutionary model discovery in complex systems.
Projects
SocialSim investigated computational approaches to modeling and forecasting online social behavior. My work centered on a multi-platform simulation framework and the Multi-Action Cascade Model, connecting behavioral assumptions about influence with scalable implementation.
Read the project ↗Current-research flagship
Different generative mechanisms can produce similar observable behavior. This work investigates how machine learning and conformal prediction can be used to identify sets of plausible behavioral rules in agent-based models rather than forcing a single overconfident explanation.
Read the project ↗Publications
2025
Garibay, I., Barham, C., Abdidizaji, S., Jayalath, C.
Advances in Artificial Intelligence Applications in Industrial and Systems Engineering (eds. W. Karwowski, V. Duffy, G. Salvendy), pp. 355–370, Wiley
A book chapter surveying future directions and applications of artificial intelligence in industrial and systems engineering contexts.
2021
Garibay, I., Oghaz, T., Yousefi, N., Mutlu, E., Schiappa, M., Scheinert, S., Anagnastopoulos, G., Bouwens, C., Fiore, S., Mantzaris, A., Murphy, J., Rand, W., Salter, A., Stanfill, M., Sukthankar, G., Baral, N., Fair, G., Gunaratne, C., Hajiakhoond, N., Jasser, J., Jayalath, C., Newton, O., Saadat, S., Seneviratna, C., Winter, R., Zhang, X.
Proceedings of the 2019 International Conference of The Computational Social Science Society of the Americas, pp. 153–169, Springer International Publishing
Presents the Deep Agent framework for studying the dynamics of information spread and evolution in social networks.
2019
Gunaratne, C., Senevirathna, C., Jayalath, C., Baral, N., Rand, W., Garibay, I.
5th International Conference on Computational Social Science
Introduces the Multi-Action Cascade Model (MACM), in which each action type exerts a distinct influence on the receiver — the founding publication behind the DARPA SocialSim simulation work.
2016
Jayalath, C., Wimalaratne, P., Karunananda, A.
International Journal of Simulation Modelling, 15(4), 585–596
Models goal selection for characters in primary groups within agent-based crowd simulations, from Chathura's undergraduate honours research.