Validation and Verification of a Neural-Symbolic Hybrid System

MSc thesis (2005–2007)

Thesis: “Validation and Verification Automatic Module of Extracted Rules of a Neural-Symbolic Hybrid System”
Research center: National Center of Research and Technological Development (CENIDET), Department of Computer Sciences

Architecture of the neural-symbolic hybrid system: symbolic rules are converted into an artificial neural network, refined with examples, converted back into rules and checked by the validation and verification module — Neural-symbolic hybrid system with the validation and verification module (4)

In my MSc thesis I developed an automatic validation and verification module for a neural-symbolic hybrid system (NSHS). As neural-symbolic hybrid systems gain acceptance, it becomes increasingly necessary to guarantee the automatic validation and verification of the knowledge they contain; in the past, these processes were carried out manually.

The project presents an enhanced Petri net model to detect and eliminate structural anomalies in the knowledge base of the NSHS. In addition, a reachability model is proposed to compare the results obtained by the system with those expected by the user.

The validation and verification method is divided into two stages:

Three phases: rule normalization, rule modeling and rule verification.
Three phases: rule modeling, dynamic modeling and evaluation of results.

The method is useful to ensure that the results of an NSHS are correct. Examples demonstrate the effectiveness of the results obtained with it.