Authors:

Michael Solmos

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Synthetic data generation is vital for increasing data availability, especially for engineering processes requiring extensive data. Valuable data transfers are necessary; however, they pose the risk of interception by adversaries. There is a gap in the process of data encryption when attackers have knowledge of the specific data properties. Generation of surrogate time-series data requires the preservation of dominant trends found within the original data as well as the stochastic properties of the associated noise. Data transfers are necessary; however, they are vulnerable to cyber-attacks and potential threats of exposure. This paper proposes a novel method of synthetic data generation that includes a deception component to mislead adversaries. The primary objective of the following techniques is to preserve the dominant trends of the original data and to introduce subtle modifications within the surrogate data sets. The process decomposes an input time series into its dominant and noisy components and generates synthetic data. Spurious correlations are introduced into the data via a "deception operator" to mislead adversaries who intercept it. The data will have identical statistical properties to the original data and additionally will remain unbeknown to adversaries. The results provide compelling evidence of the method's efficacy in protecting against cyber security adversaries.

Source:

Purdue University / 2023

Topics:

No topics listed

Co-authors:

Michael Solmos