Last year, J.P. Morgan’s Onyx team won SWIFT’s innovation hackathon with the objective to generate synthetic data while preserving data privacy.

In continuation of that, the winning team has now published a paper [full paper here] on FedSyn framework that details application of three advanced techniques for generating synthetic data sets:  Generative Adversarial Network (GAN), Federated Learning and Differential Privacy. The Onyx FedSyn solution combines emerging technologies in the areas of Artificial intelligence/Deep Neural Networks and Privacy-preserving Analytics, making it possible for multiple participants to collaborate and co-create data-driven solutions for financial institutions on a network while maintaining the privacy of their data.

The Onyx team's solution is designed to provide value to banks, financial institutions, payment networks and other entities that require privacy preservation of their customer data.

FedSyn meets blockchain

FedSyn combines synthetic data generation with privacy-preserving Federated Learning:

• It attempts to address data scarcity, data privacy, data bias and augments data-centric AI.
• Differential privacy on model parameters in Federated Learning
• Generates synthetic data even from imbalanced distribution

The Onyx team, who’ve developed Liink, J.P. Morgan’s blockchain network, established that FedSyn can delegate secured aggregation to a consortium-trusted entity in a permissioned blockchain network, such as LiinK – another step forward in the firm’s work to provide network participants with an improved experience through innovative technologies and collaboration.

Liink and collaborative use cases

“As banks and financial institutions look to develop AI based solutions, data scarcity will continue to be a challenge,” says Suresh Shetty, Distinguished Engineer and Onyx CTO. “A Federated and collaborative model to develop synthetic data over a permissioned blockchain network can benefit contributors and the consumers alike.”

“Liink is an enterprise blockchain platform for information exchange that makes payments faster, cheaper and safer for institutions,” says Sushil Raja, Global Head of Liink, Onyx by J.P. Morgan. “Data privacy is an important priority for financial institutions. This research will enable us to explore use cases on the Liink network where data privacy is paramount.”

“The Onyx team’s FedSyn solution implements a modern and emerging machine learning stack that makes it "agnostic" -- easily extendable to other business use cases requiring large amounts of data and the preservation of privacy” says Sudhir Upadhyay, of Onyx Engineering.  “FedSyn implementation further delegates its computational burden, making the solution infinitely scalable through "edge computing."

“Although the experiment in the paper was conducted with public MNIST and CIFAR10 dataset, the techniques and algorithms used in the implementation are data agnostic,” Monik R Behera concludes. “This will enable engineering artifacts developed in this exercise to extend to any use case where diversity of data is scattered and distributed across different participants who are willing to collaborate, and at all times with security and privacy.”

[This is not in production, but a continued and evolving effort towards exploring various business opportunities in privacy and collaborative computing]