Assessment of Gradient Descent Trained Rule-Fact Network Expert System Multi-Path Training Technique Performance

Abstract

The use of gradient descent training to optimize the performance of a rule-fact network expert system via updating the network’s rule weightings was previously demonstrated. Along with this, four training techniques were proposed: two used a single path for optimization and two use multiple paths. The performance of the single path techniques was previously evaluated under a variety of experimental conditions. The multiple path techniques, when compared, outperformed the single path ones; however, these techniques were not evaluated with different network types, training velocities or training levels. This paper considers the multi-path techniques under a similar variety of experimental conditions to the prior assessment of the single-path techniques and demonstrates their effectiveness under multiple operating conditions.

Deceptive Online Content Detection Using Only Message Characteristics and a Machine Learning Trained Expert System

“…In response to the issues discussed in the previous subsection, a technique was proposed that goes beyond merely XAI. This technique, machine learning trained expert systems, which was introduced in [18,19], is used for the analysis performed in this paper. The technique is fundamentally different from the traditional neural network, in structure, though it provides conceptually similar machine learning capabilities.…”

“…The experiments that are described in this paper were performed using a system derived from the one used for the experimentation presented in [18,19]. Unlike those papers, which used an ideal system (in some cases with perturbations) to train and test the gradient descent-trained expert system (as described in [20]), the work presented in this paper uses real world data from the Sentimental LIAR dataset [42] for supervised training (in place of the ideal system) and performance evaluation.…”

“…Unlike those papers, which used an ideal system (in some cases with perturbations) to train and test the gradient descent-trained expert system (as described in [20]), the work presented in this paper uses real world data from the Sentimental LIAR dataset [42] for supervised training (in place of the ideal system) and performance evaluation. However, the data storage and network implementation system used for the work herein is the same as was used in [18,19] as are the training and presentation-for-evaluation mechanisms.…”

“…This is multiplied by the velocity value and the amount of difference that is being distributed. It is computed with the equation (modified from [19]):…”

“…In [18], a neural-network-like technique for training a network whose pathways are pre-defined (and, thus, not susceptible to learning non-causal or inaccurate associations) was proposed. In [18] (and further development presented in [19]), randomly generated networks, rules, and facts were utilized to demonstrate the technique and characterize its efficacy in a generalizable manner (as described in [20]).…”