Multi-level Agent-Based Modeling
Thus, the fertile viewpoint is nothing less than the "eye" which, at one and the same time, enables us to discover and, at the same time, recognize the simple unity behind the multiplicity of the thing discovered. And, this unity is, veritably, the very breath of life that relates and animates all this multiplicity.
Yet, as the word itself suggests, a "viewpoint" implies particularity. It shows us but a single aspect of a landscape or a panorama out of a diversity of others which are equally valuable, and equally "real". It is to the degree that the complementary views of the same reality cooperate, with the increasing population of such "eyes", that one's understanding of the true nature of things advances. The more complex and rich is that reality that we wish to understand, the more the necessity that there be many "eyes" for receiving it in all its amplitude and subtlety.
Alexandre Grothendieck, Récoltes et Semailles (1983-1985) p. 41-42. Translation by Roy Lisker
Yet, as the word itself suggests, a "viewpoint" implies particularity. It shows us but a single aspect of a landscape or a panorama out of a diversity of others which are equally valuable, and equally "real". It is to the degree that the complementary views of the same reality cooperate, with the increasing population of such "eyes", that one's understanding of the true nature of things advances. The more complex and rich is that reality that we wish to understand, the more the necessity that there be many "eyes" for receiving it in all its amplitude and subtlety.
Alexandre Grothendieck, Récoltes et Semailles (1983-1985) p. 41-42. Translation by Roy Lisker
What is it?
Agent-based modeling (ABM) has been recognized as a major modeling paradigm but however, suffers from important limitations.
Bottom-up
ABM is purely bottom-up: a microscopic knowledge, i.e., related to system components, is used to construct models while a macroscopic knowledge, i.e., related to global system properties, is used to validate models.
Bidirectional relations
It is not straightforward to explicitly introduce bidirectional relations between the microscopic and macroscopic points of view or introduce new ones representing, e.g., different spatial and temporal scales or domains of interest.
Scaling
Agent-based models do not scale easily and generally require large computational resources since many agents are simulated.
Multi-level agent-based modeling (ML-ABM) aims at extending the ABM paradigm to overcome these limitations.
Definition
ML-ABM consists in integrating heterogenous agent-based models, representing complementary points of view, so called levels, of the same system.
Integration
Models integrated in a multi-level agent-based model can interact and share entities such as environments and agents. It can be viewed as a strong form of coupling.
Heterogeneity
Models integrated in a multi-level agent-based model can be based on different modeling paradigms and time representation, and represent processes at multiple scales.
ML-ABM is mainly used to solve three types of problems:
the modeling of cross-level interactions, e.g., an explicit top-down feedback control
the coupling of heterogeneous models, i.e., based on different modeling paradigms
the dynamic adaptation of the level of detail of simulations
Bibliography
During last decade, multi-level agent-based modeling has received significant increasing interest and has been applied to various domains.
A comprehensive analysis of the ML-ABM literature can be found in this working paper. Note: this article has been plagiarized 😠.
The bibliographic database (updated on the 24 Jun. 2020) is available in bib and pdf formats.
Meta-models and platforms
Generic agent-based approaches
NetLogo
NetLogo is probably the most notorious ABM programming language and simulation environment and therefore has been used in many scientific projects over the years.
Since the version 6, it supports multi-level modeling through the LevelSpace extension which allows the simulation of hierarchically coupled NetLogo models.
Since the version 6, it supports multi-level modeling through the LevelSpace extension which allows the simulation of hierarchically coupled NetLogo models.
GAMA
GAMA is an ABM platform with a dedicated modeling language, GAML, that offers multi-level capabilities. Moreover, it includes a framework to agentify emerging structures. While the notion of level does not appear explicitly in GAML, the concept of species defines attributes and behaviors of a class of same type agents and how species can be nested within each other.
IRM4MLS - SIMILAR
IRM4MLS (Influence Reaction Model for Multi-Level Simulation) is a multi-level extension of IRM4S, an ABM meta-model based on the Influence Reaction principle. SIMILAR (Simulations with Multi-Level Agents and Reactions), extends IRM4MLS using more precise terminology and simulation algorithms, as well as an implementation-oriented model for the reaction phase.
PADAWAN
PADAWAN (Pattern for Accurate Design of Agent Worlds in Agent Nests) is a multi-level ABM meta-model based on IODA (Interaction-Oriented Design of Agent simulations).
Mecsyco
Mecsyco is a simulation platform based on DEVS and AA4MM (Agent and Artifact for Multi-Modeling), a multi-modeling approach applied to ML-ABM. Levels are reified by agents that interact trough artifacts.
CRIO - Janus
CRIO (Capacity Role Interaction Organization) is an organizational meta-model dedicated to ML-ABM based on the concept of holon.
It has been implemented in the Janus platform.
It has been implemented in the Janus platform.
Platforms dedicated to biological systems
SPARK
SPARK (Simple Platform for Agent-based Representation of Knowledge) is a framework for multi-scale ABM, dedicated to biomedical research.
ML-Rules
ML-Rules is a rule-based multi-scale modeling language dedicated to cell biological systems, implemented with the framework JAMES II.
ABM-TKI
ABM-TKI is a brain tumor ML-ABM available as a MATLAB library. It is based on a 4 level architecture (tissue, microenvironmental, cellular, modelcular).