An Integrated Process for Verifying Deep Learning Classifiers Using Dataset Dissimilarity Measures

The specification and verification of algorithms is vital for safety-critical autonomous systems which incorporate deep learning elements. We propose an integrated process verifying artificial neural network (ANN) classifiers. This consists off-line on-line performance prediction phase. intended to verify ANN classifier generalisation performance, this end makes use dataset dissimilarity measures. introduce a novel measure quantifying the between used train classification algorithm, test evaluate performance. A system-level requirement could specify permitted form functional relationship measure; such be verified by dynamic testing. Experimental results, obtained using publicly available datasets, suggest that measures have relevance real-world practice both dissimilarity, specifying