Kay Chen Tan

Modeling Civil Violence: An Evolutionary Multi-Agent, Game Theoretic Approach

This paper focuses on the design and development of a spatial evolutionary multi-agent social network (EMAS) to investigate the underlying emergent macroscopic behavioral dynamics of civil violence, as a result of the microscopic local movement and game-theoretic interactions between multiple goal-oriented agents. Agents are modeled from multi-disciplinary perspectives and their behavioral strategies are evolved over time via collective co-evolution and independent learning. Experimental results reveal the onset of fascinating global emergent phenomenon as well as interesting patterns of group movement and behavioral development. Analysis of the results provides new insights into the intricate behavioral dynamics that arises in civil upheavals. Collectively, EMAS serves as a vehicle to facilitate the behavioral development of autonomous agents as well as a platform to verify the effectiveness of various violence management policies which is paramount to the mitigation of casualties.

A Coevolutionary Paradigm for Dynamic Multi-Objective Optimization

As pointed out in the previous chapter, it is imperative that the MOEA must be capable of attaining high convergence speeds in order to find the optimal solution set before it changes and becomes obsolete. However, high convergence speed often implies a rapid loss of diversity during the optimization process, which inevitably leads to the inability to track the dynamic Pareto front. Therefore, it is necessary to maintain or generate sufficient diversity to explore the search space when the multi-objective problem changes.

Dynamic Evolutionary Multi-objective Optimization

Many real-world systems include time-varying components and, very often, the environment in which they operate is in a constant state of flux. For problems involving such dynamic systems, the fitness landscape changes to reflect the time-varying requirements of the systems. Examples of such problems can be found in the areas of control, scheduling, vehicle routing, and autonomous path planning.

Robust Evolutionary Multi-objective Optimization

Branke [30] considered robust optimization as a special case of dynamic optimization, where solutions cannot be adapted fast enough to keep in pace with environmental changes. In such cases, it would be desirable to find solutions that perform reasonably well within some range of change. In fact, many real-world applications involve the simultaneous optimization of several competing objectives and are susceptible to decision or environmental parameter variations, which result in large or unacceptable performance variations. Robust optimization of multi-objective problems is the third and final type of uncertainty considered in this work and it involves the optimization of a set of Pareto optimal solutions that remains satisfactory in face of parametric variations.

Handling Noise in Evolutionary Neural Network Design

In this chapter, we consider the design of artificial neural networks (ANNs) as an instance of noisy design problem. In the context of ANN design, evolutionary optimization has led to the development of evolutionary artificial neural networks (EANN) in which adaptation is performed primarily by means of evolution. Given that the intrinsic relationship between the architecture and the associated synaptic weights can be quite complex, the design methodology would be flawed if we were to decouple these two properties during the training phase of the network. The design of ANN has two intrinsical noise sources:

Handling Noise in Evolutionary Multi-objective Optimization

In the previous chapter, we have shown empirically that the performance of MOEA deteriorates quickly with increasing noise intensities. As the results suggest, the canonical MOEA will face difficulties identifying non-dominated solutions, let alone maintaing a diverse set of near-optimal solutions.

Evolving Robust Routes

In the previous chapter, we described the VRPSD as an instance of real-world robust problem. The VRPSD differs from its deterministic counterparts in that when some data are random, it is no longer possible to require that all constraints be satisfied for all realizations of the random variables (170). In addition, the actual cost of a particular solution to the VRPSD cannot be known with certainty before the actual implementation of the solution. Optimization of the VRPSD deterministically may yield very good route schedules which we will show in this chapter to be very sensitive to variations in customer demand.

Noisy Evolutionary Multi-objective Optimization

In the previous chapter, we have described the multi-objective optimization problem and the challenges that it entails. However, the formulation presented in equation (1.1) assumes that the objectives can be found deterministically, which is hardly the case in many real world problems. Noise stems from several sources, including sensor measurement errors, incomplete simulations of computational models, and stochastic simulations. Apart from these external sources, noise can also be intrinsic to the problem. A good example is the evolution of neural networks where the same network structure can give rise to different fitness values due to different weight instantiations [144].

Evolving Robust Solutions in Multi-Objective Optimization

As described in Chapter 7, variations in design variables or the environment may affect solution quality and design performance adversely and robust optimization considers the effects explicitly and seeks to minimize the consequences without eliminating efficiency. Many different approaches, including Taguchi orthogonal arrays, response surface methodology, probabilistic design analysis, have been applied for robust optimization. In operational research, robust optimization is considered as a modeling methodology where robust problems are reformulated into the form of linear, conic quadratic, and semi-definite programming problems. Nonetheless, assumptions or approximations are often made during problem reformulation to ensure computational tractability, resulting in more uncertainties in the problem model. In addition, it does not allow for the incorporation of any domain knowledge to achieve better performance. On the other hand, evolutionary optimization techniques do not have such limitations, making it appropriate for robust optimization.