Abstract:
The development of an inductive learning agent for simulating evolutionary processes, which is meant to be linked to the ALBATROSS model, is discussed. The agent was developed to simulate learning and adaptation in travel behavior. It is assumed that at the start of the process, characterized by a high degree of uncertainty, individuals will display information-seeking behavior and choose the alternative that maximizes the entropy of outcomes. As learning proceeds, individuals will switch to goal-seeking behavior and select the alternative that maximizes the expected size of the outcome value. A series of computer experiments using simulated data were conducted to illustrate the system. Simulations showed that the system is able to learn even under conditions of noisy feedback, irrelevant attribute information, and fairly complicated interactions between relevant attributes. A combined information- and goal-seeking strategy increases learning speed.
