Frugality in reasoning and the role of summary

This paper describes a system for problem-solving in complex, dynamic environments while adhering to the principle of frugality. Given a formulation of a problem as a set of hypotheses, any such system must be able to actively search for information to confirm or refute the hypotheses. However, rather than incorporate new information under a philosophy of minimal change, we argue the system should periodically summarise its working knowledge base and adjust the priority given to alternate hypotheses. Mechanisms for forming the summary are presented, adapting ideas used in text management. Demand forecasting in volatile environments is an important application for this type of "commonsense" system. An example is provided of the performance of such a system against a standard regression forecaster on a sales forecasting task with noisy data.