Continual Neural Learning Computational Method for In-Ear based User Authentication Systems

Biometric authentication plays an important role in many emerging security solutions and as such presents an active area of research. In particular, biometric applications that employ some sort of neural artificial intelligence solution are becoming more prominent in the security and access control field. However, how best to learn the important (salient) biometric features particularly with respect to a streaming set of data makes it more challenging for existing neural techniques. This research paper presents bio-inspired neural computational technique that is of a progressive nature (learns continually) for the task of in-person prediction. The technique specifically employs an emerging Continual Neural Learning (CNL) technique to predict the authenticity of a person. For this prediction task, simulations are performed by feeding continual streams of in-ear biometric data obtained from a real world subject to the CNL technique which predicts the likelihood of a valid or invalid feature pattern. The results of simulations are reported in terms of the Mean Absolute Percentage Error (MAPE) and validated using two progressive (continual) learning approaches - an existing approach called the progressive Long Short-Term Memory (pLSTM) and an emerging approach called the Neuronal Auditory Machine Intelligence (NeuroAMI). From the results, it was found that the NeuroAMI gave best MAPE error estimates considering 3 different random data corruption trials. Hence, the NeuroAMI is recommended as a candidate CNL approach for in-ear authentication systems.