On the Effectiveness of Integrating Probabilistic Neural Networks with Influence Diagrams

論整合神經網路和影響圖的效能

曾敏烈(Ming-Lei Tseng)

《華岡工程學報》 16期 (2002/06) Pp. 27-53

https://doi.org/10.29987/CCUHKJE.200206.0002

摘要

眾所皆知，神經網路(Neural Networks)具有良好的學習能力、錯誤容忍力、以及大量的平行處理能力；但是，欠缺推論和解釋能力。本研究旨在結合影響圖(Influence Diagrams)的經濟、效率和彈性，爲神經網路提供較接近人類的推論和解釋能力。特別是，此整合網路在降低神經網路結構複雜性和推論上的複雜性(Complexity)，具有很好的效能。

並列摘要

Although neural networks indeed offer important new approaches to information processing, it is now a general trend to incorporate conventional computer methodologies for inferences and explanations in the hope of getting closer to human performance. In this study, we propose influence diagrams as a means of not only providing probabilistic inference but also reducing system complexity.

並列關鍵字

neural networks ； influence diagrams ； complexity