E.E. Vityaev and S.O. Smerdov (Russia)
Prediction, Probabilistic Inference, Probabilistic LogicProgramming
Predictions are very important for many Artificial Intelligence tasks and systems, such as expert systems, decision support systems, control systems and robotics. But prediction notion encounters with some deep problems are to be solved yet. We will consider Deductive-Nomological (D N) and Inductive-Statistical explanations/predictions. D N explanations/predictions are treated as predictions in accordance with ’The Logic of Scientific Discovery’ by K. Popper [1]. According to this work we cannot apply D N explanations/predictions to inductively obtained knowledge. We argue that logical inference of predictions from inductively obtained knowledge induces some problems relating to probability and logic synthesis. To avoid this complications we propose an inductive inference of predictions without logical inference. We will define an inductive inference of predictions (Semantic Probabilistic Inference (SPI)) and a p-prediction. For any literal A p prediction inductively infers a rule, that predicts this literal with estimation no less than the corresponding estimations obtained by probabilistic logic or probabilistic logic programming. Moreover, we prove that inductively inferred rules possess many important properties: for example, predictions based on these rules are free from the problem of statistical ambiguity. Finally, we will mention the program system ’Discovery’, implementing SPI, which was successfully applied for solution of many practical tasks (see www.math.nsc.ru/AP/ScientificDiscovery)
Important Links:
Go Back
