How Federated Learning Can Address Data Privacy Issues in banking
Enterprises that operate with large volumes of sensitive data are not always privy to the data acquisition process, which eventually leads to serious data privacy concerns. Federated Learning (FL) – a form of machine learning (ML) that trains algorithms on devices distributed across a network, without the need for data to leave each device – can be a powerful solution to address the growing concern of data privacy.
This new approach for training deep networks was first voiced in a 2016 paper published by Google AI, Communication – Efficient Learning of Deep Networks from Decentralized Data. Given the increasing awareness of privacy issues, FL could become the preferred method of ML for use cases that use sensitive data.
Take for example the BFSI sector. CIOs in banking and finance often create a distributed ML system that gives them great insights from the underlying data to enable them to make informed decisions to serve their customers as well as regulators, without worrying about data sharing, availability, and data privacy.
It’s fascinating how a distributed machine learning paradigm that enables multiple parties to jointly train a shared model without exchanging their data with any parties and uses distributed computing, AI and cryptography in conjunction to provide intelligent solutions with maximum data protection.
This collaborative yet non-data-sharing approach assures preserving sensitive data of the owner while the learnings get transferred into a single model enabling timely, informed insights management making it one of the smartest and yet most simplistic ways to train ML without creating data vulnerability.
It is important to note that when it comes to transaction validation, fraud detection, credit risk, etc., in the financial services industry FL often addresses the grey areas, where data privacy may be at stake, and where bankers refrain themselves to put certain AI models in use to preserve customer confidentiality.
By automating numerous tasks, this technique can help banks in lowering the operational cost. FL holds its credibility in recognizing fraudulent transactions, malicious activities and enhancing user experience. One of the most dependent features of FL is to provide a secure ecosystem for users and bankers. In fact, it is proved to be far more secure when compared to traditional ML, and its encryptions to model offer an additional layer of security.
Unsurprisingly, the popularity and implementation of FL have spiked, and many new frameworks, backed by tech-leading companies are showing confidence in this technology.
As with implementation, researchers are also working tirelessly to make improvements in the orchestration of FL. They are striving to find the optimal way to aggregate the local models and though pretty nascent it seems we are almost there.
Undoubtedly, FL is the future of privacy preservation and AI model training, and the scope of this secured distributed ML is immense. Banks, healthcare and applications we use on our devices, all hold enough data to train a highly accurate and efficient AI engine.
As FL is evolving, and the goal is to achieve 100% accuracy, Google Voice Assistance, Google Keyboard, Apple Siri, iOS update, and optimization decisions have employed this technique and it is already a success; this is enough silver lining for a new chapter in data privacy.
(The author Aditya Dangi is a Machine Learning specialist at Synechron)
