Institutional Repository [SANDBOX]
Technical University of Crete
EN  |  EL

Search

Browse

My Space

Large Differentially Private Data Synthesis

Zacharioudakis Christos

Full record


URI: http://purl.tuc.gr/dl/dias/B3CEC19D-7593-4E6F-97EE-591C237D6B63
Year 2020
Type of Item Diploma Work
License
Details
Bibliographic Citation Zacharioudakis Christos, "Large Differentially Private Data Synthesis", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2020 https://doi.org/10.26233/heallink.tuc.84556
Appears in Collections

Summary

In our days, data exists in abundance, it is ever increasing and it finds numerous uses. A most recent use is the training of Machine Learning models, software capable of making their own decisions. However, using data to train said models raises significant privacy concerns, especially when it comes to highly sensitive data such as medical records. A solution to this predicament is the synthetic data generation, the production of “fake” data that resembles the real one. However, synthetic data generation does not provide any privacy guarantees on its own. The need increases for a robust, meaningful, and mathematically rigorous definition of privacy, together with a computationally rich class of algorithms that satisfy this definition. One such definition is Differential Privacy. This thesis attempts to combine the concept of Differential Privacy with various Machine Learning techniques to generate truly private data that can be utilized in place of the real one effectively. The Machine Learning models that will concern us are the Bayesian Networks and the Generative Adversarial Networks.

Available Files

Services

Statistics