Validating Siamese embedded neural networks with identical representations for efficient model convergence

Mathias Hoy Talbo, Haishuai Wang, Lianhua Chi, Yi-Ping Phoebe Chen

Abstract: Highlights•Generate valid Siamese embedded neural network architectures with identical pairs.•Produce efficient converging models with transform learning on same-class pairs.•Reduce hyperspace with our validation technique and modified tangent activation function.•Ascertain similarity independent of spatial placement with our non-metric asymmetric function.•Train pairs with mask for consist transformation and evaluate without to minimise influence.