RTGAN: An Novel GAN-Based Rule Transfer Learning Method for Scalable Industrial Inference Engine

Nowadays deterministic inference based on rules is playing an increasingly important role in industrial applications. Compared with non-deterministic inference, it is fast, accurate and interpretable. However, the current rule-based inference engine costs a large amount of manpower and lacks of ability to adapt to production line changes and version changes. Therefore, combined with the idea of analogical reasoning of human being, we propose RTGAN, a novel GAN-based rule transfer learning method for scalable industrial inference engine. RTGAN constructs generators using transfer domain knowledge graphs and explicit representation of analogous relationships using an attention-based mechanism of the embedding graph matching algorithm. The RTGAN based inference engine will be implemented and adequate experimentation will be carried out in large-scale rule sets.