A Hybrid Autoencoder Architecture for Text Encryption

Abstract

With the advent of modern computing technologies, there is a wide scope for introduction of advanced and highly secure encryption standards that cannot be rendered obsolete even when subjected to these techniques. This paper introduces one such encryption paradigm, implemented using autoencoders. The proposed algorithm trains an autoencoder to replicate an input 8-bit ASCII value and takes advantage of the bottle-neck architecture to encrypt the data in a way that can only be retrieved using the trained weights. Various analyses are utilized to exemplify the speed as well as security of the proposed cryptosystem.