6. RESULTADOS Y DISCUSIÓN
6.1 Regulación periférica y central de la ingesta y la saciedad
6.1.2 Regulación central y comunicación con la periferia
With the rapid growth of computational capability, simulation has become a powerful tool to investigate complex systems in animals (DeAngelis & Grimm, 2013). These ecological simulations are typically agent-based and spatial-explicit models, which hold
44
a population of agents in a virtual arena and mimic the real-world system by spatiotemporal interactions among these agents (Couzin & Krause, 2003; Schellinck & White, 2011). The advantage of this kind of simulation is that the abstract formulations are largely avoided compared with traditional theoretical models (Schellinck & White, 2011; DeAngelis & Gimm, 2013). Therefore, the implicit causality of a natural system is inferred more directly and clearly.
One of the most successful usages of simulation is to reveal the self-organising mechanism of group-living fish’s collective movement (Schellinck & White, 2011). For example, Aoki (1982) demonstrated that fish schools can be an emergent pattern by agents of simple behaviours. Huth & Wissel (1994) compared various behavioural settings and found that fish’s schooling pattern is better mimicked when decision making relies on multiple neighbours. Couzin et al. (2002) showed that the transition between fish’s collective patterns can be displayed by adjusting a certain behavioural preference of agents. These simulations have largely improved the understanding of fish’s behavioural mechanism (Couzin & Krause, 2003).
After the success of modelling fish’s behaviours by computers, simulation has recently been employed to study the behavioural evolution of fish (Sumpter, 2006). That is to say, why should fish develop these simple behaviours in evolution? However, at present, simulations on this topic are less significant due to the diverse outputs from different designs. For example, Wood & Ackland (2007) demonstrated that the survival pressure can drive fish agents into one of the two states: a milling herd or a moving school. On the contrary, schooling was the only evolutionary result in Oboshi et al. (2003) and Ioannou et al. (2012). On the other hand, Olson et al. (2013) demonstrated
45
that schooling should be an intermediate phase in evolution and should be replaced by milling herds and stationary swarms.
The inconsistency among the previous evolutionary simulations highlights the issue of preconceived biases in ecological modelling. For example, if the work in Wood & Ackland (2007) is considered more successful due to its ‘reasonable’ output (Sumpter, 2009), the validation then biases to the preconceived knowledge and reserves no significance for the experiments by simulation. Unfortunately, validation in ecological modelling has been highly contested (Rohani et al., 1997; Parrish et al., 2002; Schellinck & White, 2011), and a convincing metric to score ecological simulations still seems impractical. Nevertheless, two fundamental principles have been widely agreed, which are realistic settings and reliable outputs (Grimm et al., 2005; Schellinck & White, 2011).
The most important factor of a valid ecological simulation is that the setting should accord with the empirical data (Parrish et al., 2002; Grimm et al., 2005; Schellinck & White, 2011). Otherwise, the output cannot be an analogy to the targeted natural system. For example, although fish schools can be mimicked vividly by averaging the influences of all neighbours (Huth & Wissel, 1994; Couzin et al., 2002), this setting has become disputed after unfavourable empirical evidence was found (Ballerini et al., 2008; Katz et al., 2011). From this viewpoint, the model in Wood & Ackland (2007) may not be as valid because at least two of its settings violate the empirical data considerably. One is that an agent’s decision making in this model is also evenly influenced by all neighbours. Another is that the paralleling behaviour of its agents is not observed in the real fish (Tien et al., 2004; Katz et al., 2011). Comparatively, Olson et al. (2013) designed the movement of agents without predetermined rules. This design
46
is more realistic to a certain extent, despite its output, as the crucial part of a model, may be unrealistic.
The credibility of a model is increased if its outputs are less sensitive to potential noises and parameter adjustments (Rohani et al., 1997; Grimm et al., 2005), which property is also referred to as the ‘robustness’ of a demonstration (Grimm et al., 2005). Related techniques to improve an evolutionary model’s ability to reproduce consistent and reliable outputs have been developed for decades in Evolutionary Computation (Haupt & Haupt, 2004). However, rare ecological models have acquired this knowledge. For example, Wood & Ackland (2007) designed an unscaled proportional fitness selection in its genetic algorithm, which scheme has been proven less robust than others (Whitley, 1989; Blickle & Thiele, 1995; Reeves, 2003; Noraini & Geraghty, 2011). Moreover, almost none of the previous works reported quantitative analyses about the robustness of their simulations and the sensitivity of their parameter settings, the neglect of which may cause concern about the reliability of the outputs.
In this chapter, an evolutionary model is built to reduce preconceived biases in previous works and its validation is emphasised. After building the model based on an open- ended solution space in Chapter 3.2, the validity of this model is highlighted from the three aspects. First, for an objective description of the simulation outputs, the quantification is discussed in Chapter 3.3. Secondly, to validate the authenticity of this model, a comprehensive comparison of the model, the empirical data and previous works is reported in Chapter 3.4. Thirdly, in Chapter 3.5, the parameters are scanned to understand their influences to the simulation. These elaborate analyses construct a solid base of the validity of the model outputs and further inferences.
47