OBSERVER: A Probabilistic Learning System for Ordered Events

Given a sequence of observed events which are ordered with respect to time or positions and are described by the coexistence of several discrete-valued attributes that are assumed to be generated by a random process, the inductive prediction problem is to find the probabilistic patterns that characterize the random process, thereby, allowing future events to be predicted. This paper presents a probabilistic inference technique for solving such a problem. Based on it, a learning program called the OBSERVER has been implemented. The OBSERVER can learn, inductively and without supervision, even if some observed events could be erroneous, occasionally missing, or subject to certain degrees of uncertainty. It is able to reveal the patterns and regularities inherent in a sequence of observed events and can not only specify, in a clearly defined way, the happenings in the past but also gain insight for prediction. The proposed technique can be applied to solve different problems in artificial intelligence (AI) and pattern recognition (PR) where decisions concerning the future have to be made.