« 3.4 Game Theory | HomePage | 3.5 Cellular Automata »
Friday, June 03, 20051117776300
3.6 Neural Networks
Note: This is an excerpt from a draft of my thesis, A Computer Model of National Behavior. The introduction and table of contents are also available
3.6 Neural Networks
Clark describes neural networks as models of cognition based on the neurons in the human brain. In this model there are many layers of neurons which can affect each other through forward and backward propagation. A neuron receives inputs through dendrites, decides whether or not to pass a signal by comparing its inputs to a threshold in the nucleus, outputs through its synapses. Neural nets are most appropriate when no clear algorithm is available to perform the task, and when the model will need to adapt itself to new situations quickly.
Figure 6. Neuron with Component Parts
Neural networks have been shown to explain the behavior of insects, animals, and in some areas humans, so perhaps they might explain the behavior of national populations of humans. Because there are many neurons involved, the calculations will be necessarily parallel, opening the door to optimizations. Additionally, the proven ability for biological neural networks to learn concepts and plot strategies have obvious benefits for nations.
However, despite these advantages, neural networks for modeling nations' behavior.
First, a neural model would lose explanatory power. One of the potential justifications for the model of this thesis is that it can be used as a teaching aid in explaining how nations behave. However, a neural model would result in a method that would be too complex for students to understand. Any attempt to give a summation of the model would degenerate into a meaningless listing of weights
Relatedly, a solution that presumes that there is no clear algorithm is not wanted. Like a wide variety of other political science models, such as those by Gamble, Johnston, and Olsen, the model adopted in the thesis should be deterministic and attempt to explain behavior with reference to simple rules. Therefore, even if a neural net model produced the correct results, the system will not achieve its goal.
Finally, one of the major advantages of the neural net model is not needed. The system will not have to deal with suddenly different situations. For instance, neural nets are well suited to determining if two shots from two angles show the same person. The network can handle the abrupt change in perspective. But the ebb and flow of nations is far smoother. The same small number of rules are always working on nations, and only relatively few values are changing. This situation is different from image recognition, where an astonishing set of factors have to be considered. Whereas the first two objections show neural nets to be deficient, the final objection furthers the point that a neural network is simply inappropriate for the task at hand.
Computer Science Thesis Index
00:25 Posted by Dan tdaxp in Thesis | Permalink | Comments (0) | Email this