Privacy Preserving Machine
Learning Techniques
Amogh Tarcar
Machine Learning Researcher
Index
• Need for Privacy Aware Machine Learning
• Federated Learning Intro
• FL in healthcare
• Privacy concerns
• Tools & Platforms
• Demo
Need for Privacy Aware Machine Learning
• Data sources such as EHR , X-Rays, Genomics data are inherently
sensitive and private, and there are ethical as well as legal limitations
for aggregating sensitive data
• Healthcare datasets , especially for rare diseases needs collaboration
of health providers across the world
• Privacy aware machine learning technique enables building models
without needing data to move from its source location. One such
technique is called Federated Learning.
Federated
Learning
Source : https://blog.fastforwardlabs.com/images/editor_uploads/2018-10-31-181344-federated_learning_animated_labeled.gif
Federated
Machine
Learning for
Healthcare
Source: https://blogs.nvidia.com/wp-content/uploads/2019/10/federated_learning_animation_still_white.png
Privacy Concerns
• Federated learning is a variant of
decentralized machine learning
• Even though data does not leave the
source location, the model parameters
may leak info about the data
• Federated learning needs to be
implemented using secure techniques
such as :
• Differential Privacy / Secure
Aggregation
• Secure Multi Party Computation /
Encryption
Image Source : https://www.google.com/url?sa=i&url=http%3A%2F%2Fblog.fastforwardlabs.com%2F2018%2F11%2F14%2Ffederated-
learning.html&psig=AOvVaw0tpZYKhKDMX3h5MOCEXkNC&ust=1585489636330000&source=images&cd=vfe&ved=0CAIQjRxqFwoTCLiGuqmnvegCFQAAAAAdAAAAABAD
Tools & Platforms
TensorFlow FederatedPySyft
NVIDIA CLARA
FedAI
OWKIN Platform
Healthcare Use case Demo

Privacy preserving machine_learning_current_landscape