Figure - available from: Neural Computing and Applications
This content is subject to copyright. Terms and conditions apply.
Source publication
Parameter calibration is a significant challenge in agent-based modelling and simulation (ABMS). An agent-based model’s (ABM) complexity grows as the number of parameters required to be calibrated increases. This parameter expansion leads to the ABMS equivalent of the "curse of dimensionality". In particular, infeasible computational requirements s...
Citations
... Surrogate methods have been extensively applied in single objective optimisation [36][37][38][39][40][41][42][43], whereas the applications of surrogate-assisted methods within a MO setting is substantially less common. In particular, the literature for datadriven evolutionary MO surrogate-assisted optimisation is especially sparse [44][45][46][47]. ...
Portfolio management is a multi-period multi-objective optimisation problem subject to various constraints. However, portfolio management is treated as a single-period problem partly due to the computationally burdensome hyper-parameter search procedure needed to construct a multi-period Pareto frontier. This study presents the Pareto driven surrogate (ParDen-Sur) modelling framework to efficiently perform the required hyper-parameter search. ParDen-Sur extends previous surrogate frameworks by including a reservoir sampling-based look-ahead mechanism for offspring generation in evolutionary algorithms (EAs) alongside the traditional acceptance sampling scheme. We evaluate this framework against, and in conjunction with, several seminal multi-objective (MO) EAs on two datasets for both the single- and multi-period use cases. When considering hypervolume ParDen-Sur improves marginally (0.8%) over the state-of-the-art (SOTA)-NSGA-II. However, for generational distance plus and inverted generational distance plus, these improvements over the SOTA are 19.4% and 66.5%, respectively. When considering the average number of evaluations and generations to reach a 99% success rate, ParDen-Sur is shown to be 1.84× and 2.02× more effective than the SOTA. This improvement is statistically significant for the Pareto frontiers, across multiple EAs, for both datasets and use cases.
... Calibration in ABM is a serious challenge (Challenge III) especially as the complexity grows with increasing the required number of petameters subject to calibration (Perumal & Zyl, 2022). ML could be beneficial in this step in various ways. ...
In recent years, computationally intensive theory construction, leveraging big data and machine learning (ML), has gained significant interest in the information systems (IS) community. The integration of computational methods can generate novel methodological paradigms or enhance existing methods. Agent-based modeling (ABM) is one of the computational methods that has recently proliferated in IS research to generate computationally intensive theories. However, ABM is still in nascent state of adoption in IS research and entails some pathological challenges that limit its applicability and robustness. With the goal of advancing ABM in IS research, this article proposes a methodological framework that integrates ML within relevant steps of ABM. The framework is demonstrated in an exemplary IS study, showing its potential for addressing the pathological challenges of ABM. We finally discuss the implications of applying the proposed methodological framework in IS research.
... Marzouk and Hassan (2022) employ the k-means algorithm to cluster the evacuation performance of both museum visitors and workers in response to fire. The supervised ML algorithms used for validation purposes include genetic algorithms White et al., 2022), Boost DT (Lamperti et al., 2018;Zhang et al., 2020), random forest (Pawar & Jha, 2022), support vector machines (Perumal & Zyl, 2022), simulated annealing (Neri, 2019), neural networks (Jørgensen et al., 2022;Krivorotko et al., 2022;Reiker et al., 2021;van der Hoog, 2019), and Bayesian networks (Chu et al., 2018;Masuda et al., 2022). It is worth noting that the interpretability of the algorithm is not relevant in this case. ...
Agent‐based models have diversified their applications across various domains due to the ease with which different phenomena can be represented and simulated. These models incorporate heterogeneous, autonomous agents, local interactions, bounded rationality, and often feature explicit spatial representations. However, certain challenges have been identified in their application, including the complexity of design, difficulty in calibrating parameters, and interpreting and analysing results. Therefore, incorporating machine learning (ML) tools in the various stages of the agent‐based modelling and simulation process presents a promising approach for current and future research. The main hypothesis of this study is that integrating ML techniques and tools into agent‐based modelling can help address challenges encountered during different stages of implementation, ultimately leading to more accurate and effective simulations. The methodology employed in this study involves a comprehensive search and analysis of relevant literature on the topic. This survey reviews significant developments in the integration of ML into the agent‐based modelling and simulation process in recent years. The results of this study summarize the fundamental concepts of ML and its applications in agent‐based modelling, and provide insights into the prospects and challenges for ML‐assisted agent‐based modelling in the near future.
... Surrogate methods have been extensively applied in single objective optimisation [32][33][34][35][36][37][38][39], whereas the applications of surrogate assisted methods within a MO setting are substantially less common. In particular, the literature for data-driven evolutionary MO surrogate assisted optimisation is especially sparse [40][41][42][43]. ...
p>Portfolio management is a multi-period multi-objective optimization problem subject to a wide range of constraints. However, in practice, portfolio management is treated as a single-period problem partly due to the computationally burdensome hyper-parameter search procedure needed to construct a multi-period Pareto frontier. This study presents the Pareto Driven Surrogate (ParDen-Sur) modelling framework to efficiently perform the required hyper-parameter search. ParDen-Sur extends previous surrogate frameworks by including a reservoir sampling-based look-ahead mechanism for offspring generation in Evolutionary Algorithms (EAs) alongside the traditional acceptance sampling scheme. We evaluate this framework against, and in conjunction with, several seminal Multi-Objective (MO) EAs on two datasets for both the single- and multi-period use cases. Our results show that ParDen-Sur can speed up the exploration for optimal hyper-parameters by almost 2x with a statistically significant improvement of the Pareto frontiers, across multiple EAs, for both datasets and use cases.</p
... Surrogate methods have been extensively applied in single objective optimisation [32][33][34][35][36][37][38][39], whereas the applications of surrogate assisted methods within a MO setting are substantially less common. In particular, the literature for data-driven evolutionary MO surrogate assisted optimisation is especially sparse [40][41][42][43]. ...
p>Portfolio management is a multi-period multi-objective optimization problem subject to a wide range of constraints. However, in practice, portfolio management is treated as a single-period problem partly due to the computationally burdensome hyper-parameter search procedure needed to construct a multi-period Pareto frontier. This study presents the Pareto Driven Surrogate (ParDen-Sur) modelling framework to efficiently perform the required hyper-parameter search. ParDen-Sur extends previous surrogate frameworks by including a reservoir sampling-based look-ahead mechanism for offspring generation in Evolutionary Algorithms (EAs) alongside the traditional acceptance sampling scheme. We evaluate this framework against, and in conjunction with, several seminal Multi-Objective (MO) EAs on two datasets for both the single- and multi-period use cases. Our results show that ParDen-Sur can speed up the exploration for optimal hyper-parameters by almost 2x with a statistically significant improvement of the Pareto frontiers, across multiple EAs, for both datasets and use cases.</p
Agent‐based modelling has gained recognition in the last years because it provides a natural way to explore the behaviour of social systems. However, agent‐based models usually have a considerable number of parameters that make it computationally prohibitive to explore the complete space of parameter combinations. A promising approach to overcome the computational constraints of agent‐based models is the use of machine learning‐based surrogates or metamodels, which can be used as efficient proxies of the original agent‐based model. As the use of metamodels of agent‐based simulations is still an incipient area of research, there are no guidelines on which algorithms are the most suitable candidates. In order to contribute to filling this gap, we conduct here a systematic comparative analysis to evaluate different machine learning‐based approaches to agent‐based model surrogation. A key innovation of our work is the focus on classification methods for categorical metamodeling, which is highly relevant because agent‐based simulations are very often validated in a qualitative way. To analyse the performance of the classifiers we use three types of indicators—measures of correctness, efficiency, and robustness—and compare their results for different datasets and sample sizes using an agent‐based artificial market as a case study.
