Structure optimization of prior-knowledge-guided neural networks

Prior-knowledge use in neural networks, for example, knowledge of a physical system, allows network training to be tailored specific problems. Literature shows that prior-knowledge enhances predictive performance. Research date focuses on parametric optimization rather than structure optimization. We present new framework optimize the using prior-knowledge. This is achieved through optimizing number hidden units via line search and cross-validation empirical error eliminate data-set/model-structure application dependency guided networks. In addition model step, we propose utilizing prior errors as part performance index improve generalization. Results demonstrate proposed model’s prediction accuracy consistency convex data sets with unique minimum non-convex multi-modal sets. The presented results yield understanding physics-guided networks terms their structural