MON 2:00 PM Parallel - System Dynamics and Agent Based Methods

Full Report:
Representing Progression and Interactions of Comorbidities in Aggregate and Individual-Based Systems Models, by Nathaniel Osgood
This work seeks to highlight how co-morbidities, important sources of heterogeneity in health and epidemiological models, can be handled in dynamic models. In so doing, it shows the challenges that can be introduced in trying to handle such heterogeneity in aggregate versus individual-based models. Co-morbidities are defined as the presence of multiple diseases or disorders at the same time that can be related to common risk factors and also have interacting unidirectional and bidirectional causality. This complexity is important for understanding disease treatment and mitigation but at the same time introduces a significant challenge for modeling. Osgood identifies some desired characteristics of a model for co-morbidities including modularity, transparency, expressiveness, computational efficiency, and flexibility, then explores how different model schemes (aggregate versus individual) address these. The advantages of aggregate models such as mean linkage, parallel flows and attribute disaggregation are their simplicity. There is a trade-off of handling either interventions that target combinations of co-morbidities in the former two versus the lack of modularity or transparency in the attribute disaggregation case. 

Individual-based representations explored include state equations, state charts and hybrid automata. The individual-based models in general accommodated the desired characteristics of accuracy, modularity, expressiveness, and transparency. Only hybrid automata, which use state equations and a discrete state transition in a single person, also provided the desired flexibility. Future work will explore the creation of modeling language for handling co-morbidities along the lines of hybrid automata that meet all the desired criteria. Questions from the audience addressed the applicability of such models to actual diagnosis and clinical decision-making. Osgood answered that the research is still a long way from being applied in practice. Other questions regarded the applicability of co-morbidity work to the field of human resources and Osgood felt that there were potential parallels between the two.

Individual versus Group Rationality: A Co-evolutionary Approach to the Beer Game, by Hongliang Liu, Enda Howley, Jim Duggan
The beer distribution game is a hallmark for demonstrating the value of dynamic models for understanding human decision making under uncertainty and potential lessons for successful management. This paper explores how the results of the beer game may be affected by scenarios in which individual actor decisions are taken into account rather than at the aggregate level at each point in the supply chain. To explore this subject, Liu and his colleagues implore a co-evolutionary approach that uses genetic algorithms as a base which reproduces, recombines, mutates, and selects in order to create an evolution of fitness strategies as game play continues.

Each sector (retailer, wholesaler, distributor, and factory) is populated by a set of agents who can independently evolve within each sector. Overall performance for each sector is an aggregate for the population. Two schemes of rationality are then tested in terms of the fitness functions: either individual (individual cost is minimized regardless of whole supply chain cost) or global (all agents are rewarded based on entire supply chain performance). Overall, global fitness functions result in lower global costs, but retailers and wholesalers, downstream sectors, can perform better under individual fitness functions while factories and distributors perform worse. Future work will include looking at more complex customer demand functions. Questions from the session addressed the topic of hybrid co-evolutionary approaches such as the use of individual and global fitness functions for different sectors in the same game which have not yet been addressed; a participant recommended the authors explore the work of Axelrod.

Modeling Strategic Technology Management with a Hybrid Model, by Samuli Kortelainen, and Lauri Lattila
This work looks at technology management strategy through the use of a hybrid model of individual-based firms, customer requirements, technology, and products in a dynamic model (through the use of Anylogic software). The work attempts to combine the advantages of system dynamics (a thorough background in literature, theories, and applications) with those of agent-based models (flexibility and emergence). The resulting model is a network of firms and associated technologies and products that evolve over time in a competitive framework and in response to customer requirements. The model emphasizes the strategy of firm agility but further work is needed in the area as the modeling tools for such a complex model are still in need of development. Questions for the session explored the number of firms and markets for the model, which were ten and twenty, respectively. An added comment indicated that there is a sensitivity analysis bottleneck with such models to date due to model complexity and computational limits.

Katherine Dykes, PhD Candidate, MIT Engineering Systems Division