Scientific discovery: computational explorations of the creative process
Scientific discovery: computational explorations of the creative process
Probabilistic Horn abduction and Bayesian networks
Artificial Intelligence
Space-efficient inference in dynamic probabilistic networks
IJCAI'97 Proceedings of the Fifteenth international joint conference on Artifical intelligence - Volume 2
| Hi-index | 0.00 |
We or computers cannot have perfect knowledge about real events. Even in a very simple case, it is very difficult to have knowledge including every event which may have caused observed events. This makes it difficult to have the computer infer significant causes of observed events. However, the computer inference may supply good hints for humans, because unexpected relations detected between known factors by a computer suggest unknown events to humans, being combined with the vast human knowledge acquired by rich experience. In this paper, we aim at making the computer express "unknown" hidden causes, occurring at the observed time but not included in the given knowledge. This is for discovering temporary events, not acquiring permanent knowledge which generalizes observed data. Our method is to have the computer infer known causes of time-series of observed events, for detecting unexpectedly strong co-occurrences among known events. Then the detected relations are expressed to humans, which make significant unknown causal events easily understood. The Cost-based Cooperation of Multi-Abducers (CCMA), which we employ for the inference of causes, contributes for our purpose of discovery because CCMA can infer well about non-modelable time-series involving unknown causes.