Federated & Encrypted Learning to protect Data Privacy in Deep Learning

Darshak Shah (~darshak55)


Two of the most important areas in Software industry today are Data Science and Privacy. Many startups with great product ideas never get off the ground due to lack of training data, which companies are unwilling to share because of privacy and competitive reasons. Let alone startups, analytics departments in some large corporations have to cut through a lot of red tape to get access to the data they need to do their job.

Similarly, government agencies sometimes are not able to take full advantage of the data they have because they can not share citizen's data with third party due to privacy reasons.

One of the possible solutions used by companies is to annonymize the data. However incidents in the past have proved that clever data scientists are able to come around this and are able to get complete identity from the data.

We will first discuss about Differential Privacy and how that is applied to Deep Learning to evaluate and make sure the privacy of individuals is protected.

We will discuss two advanced Deep Learning methods that can help circumvent the issue of privacy without compromising it. Federate Learning :Training the Models locally, where the data resides and then combining the results in a way that does not transfer the data Encrypted Learning: Generating an encrypted prediction with an encrypted Neural Network trained on encrypted data-set.

With these methods, companies, governments and society at large can take advantage of advancements in Data Science, without compromising the privacy of individuals.


Basic knowledge of Machine Learning techniques

Speaker Info:

Darshak has a total of 13+ years of industry experience, with 7 yeasr experience in Machine Learning, Deep Learning and Analytics. He has built Analytics and Machine Learning departments from ground up in 3 differente organizations. He is currently the founder and CEO of Jash Data Sciences, specializing in Deep Learning and Machine Learning solutions. He has an MBA from Babson college in Boston, #1 school for Entrepreneurship.

Speaker Links:




Id: 1325
Section: Data Science, Machine Learning and AI
Type: Talks
Target Audience: Intermediate
Last Updated: