Impact of Loss Function Using M-LSTM Classifier for Sequence Data

1K. Sahityabhilash and S. Prayla Shyry

163 Views
47 Downloads
Abstract:

The increasing dependence on information systems has opened a lot of possibilities in solving real life problems and led to the increase of threat to privacy, integrity and authentication. Even though a lot of key based authentication systems are in use biometrics provide a better performance, apart from physiological biometrics like iris, thumb impression etc., For verifying a person, a behavioural biometric technique called Keystroke dynamics can be used. Biometric based user authentication is a sequence classification task. This study provides a comparison of different loss functions and their performance on keystroke dynamics data. This work uses Long Short-Term Memory (LSTM) representing Neural Network and we have taken 5 different loss functions for the study.

Keywords:

Long Short-Term Memory (LSTM), Loss Function, Cross-Entropy, Hinge, Normalisation, Biometric, Key Stroke Dynamics (KSD).

Paper Details
Month4
Year2020
Volume24
IssueIssue 5
Pages3487-3494