A Deep Learning Approach for a Symmetric Key Cryptography System.
Quinga Socasi, Francisco Jesús · 2020
In today’s digital world, immersed into tons of gigabytes coming from online transactions con- stantly flowing over the Internet, information security has become a necessity and subject of great interest. In this sense, computer scientists are highly encouraged to protect information against malicious people, who evolves fast with technology. The main purpose of Cryptogra- phy is to apply complex mathematics and logic to develop encryption and decryption processes, whereby the information is made unintelligible, and the original data is recovered, respectively. Thus avoiding unauthorized access to information. Up to date, various cryptography algo- rithms have been developed. Such as AES, 3DES, and RSA, where each cipher entails the advantages and drawbacks thereof. On the one hand, traditional cryptography ciphers apply ex- haustive serial operations using complex formulas and huge prime numbers, making encryption and decryption computing consuming and somehow vulnerable. On the other hand, recently machine learning and artificial intelligence have achieved significant improvements in Cryptog- raphy. This work proposes an alternative deep learning encryption system with two principal components: (1) A particular type of neural network called autoencoder for encryption and de- cryption of data, (2) A key generation algorithm which transforms an alphanumeric password into a sequence of integer numbers used during the random processes of the training phase. Ex- perimental results show that the proposed system overcome AES, DES, 3DES, and RSA when encrypting files of size no longer than 868KB. We also show that the proposed system may represent a meaningful contribution to the field of data security.