Cyber threats and data breaches are becoming increasingly prevalent, particularly in financial services. The banking sector is constantly on the lookout for innovative solutions to protect both institutional assets and customer information. One of the most promising avenues of exploration is the integration of Artificial Intelligence to enhance security measures, specifically in fraud detection. Fintech Review asked a few questions to Ittai Dayan, Co-founder and CEO of Rhino Health.
Tell us more about the Rhino Federated Computing Platform. What is your elevator pitch?
Rhino’s Federated Computing Platform (FCP) makes it possible for data scientists and AI developers to use data without ever taking possession of the data. We offer “Federated Computing” (FC), which is how we define the combination of Edge Computing (EC) and Federated Learning (FL). These technologies streamline data collaboration by leaving data ‘at rest’ behind data owners firewalls. This means financial institutions (and their regulators) can unlock the value of previously siloed data. It means querying, analyzing, fitting predictive models and deploying them on previously inaccessible data. All while preserving data sovereignty and reducing the expense associated with data storage & egress.
What is your background and the story behind the company?
I started my career as a physician and researcher, and Rhino was born out of the desire to unlock data silos in healthcare. I practiced medicine before becoming a management consultant and manager at the Boston Consulting Group (BCG). After ‘getting a taste’ for what digital transformation could do to healthcare at BCG, I decided to ‘get my hands dirty’ and decided to transition to the world of academic medicine again, by leading development and deployment of AI in an Academic Medical Center. I led the Mass General Brigham (MGB) Center for Clinical Data Science. It was in that role that I saw the power of FL firsthand, leading a consortium with Nvidia and 20 international institutions to complete an FL project in weeks that would have taken years under the old ‘data centralization’ paradigm.
After leaving MGB, my co-founder and I started Rhino with the idea that AI developers using sensitive data need an enterprise-ready solution to streamline the end-to-end workflow of pre-processing & harmonizing data; running queries on datasets distributed across multiple organizations, geographies, or IT systems; and ultimately training and deploying predictive models. Healthcare was a perfect proving ground as a highly-regulated and conservative industry. From there, we are now offering proven, extensible technology to developers across industries who need to work with sensitive data – be it in financial services, the public sector, etc.
What is federated learning? How does it relate to edge computing?
Simply put, FL is the technique of training a model on decentralized data. In its essence, FL is a distributed machine learning (ML) technique where multiple collaborators each have their own data, and different ML models are trained by aggregating parameters from these participants without sharing the raw data. FL enables many computational processes to identify and learn from patterns in data that remains disaggregated, including more ‘classical’ methods such as clustering and onto sophisticated deep-learning models. The local device retains the data, and the central server receives only the model parameters for aggregation and updating. This approach enables organizations to preserve data sovereignty while still maximizing the potential of their models. FL can improve ML models by enabling the use of larger and more diverse datasets, which is important for generalizability.
Edge Computing is, more broadly, the processing of data close to the source rather than centralizing the data and then processing them. FL is a specific application of edge computing. A critical step for AI development is making sure you have the right data and in the right format. We power multiple applications that enable the analysis and preparation of data. The FCP seamlessly combines these two powerful technologies in an end-to-end solution, mitigating a frequent stumbling block to ‘DIY’ solutions that are based on open-source, non-supported frameworks.
What are the potential applications for financial institutions
There are many uses for federated computing across financial services. Some that we’re exploring with partners, include:
Credit Scoring: build smarter credit scoring models while keeping sensitive customer data on individual banks’ servers. This would allow banks to benefit from a more comprehensive model without sharing raw customer data, providing credit to more customers without exposing the bank to a heightened level of risk.
Fraud Detection: Financial institutions could develop fraud detection models collaboratively, using data from various sources, without sharing specific transaction details that might violate privacy regulations.
Anti-Money Laundering: identify patterns of money laundering across multiple institutions while preserving the privacy of individual transactions and account details.
Risk Management: create risk assessment models that incorporate data from different institutions, aiding in portfolio management and risk mitigation – both for individual institutions as well as regulators.
Trade Settlement: federated computing could be used to streamline the process of settling trades, which is a complex, often manual multi-party process that could be streamlined by running AI-enabled pre-trade validation processes.
What are your predictions for the conundrum of fighting fraud whilst respecting data privacy in the years to come?
AI is the only scalable solution to keep up with the growth in and evolution of financial transactions around the world. Financial institutions and law enforcement agencies have to consider not only traditional payments but also trillions of dollars changing hands via ecommerce, mobile payments, and cryptocurrency. Policymakers are simultaneously wrestling with how to unleash the potential of AI while balancing the privacy concerns of their citizens.
FC is the answer. Financial institutions and regulators are increasingly realizing that FC offers the elegant solution of both allowing for robust fraud detection models using data from disparate silos while also preserving customer privacy and complying with relevant regulation.
Any other innovation in fintech elsewhere that you are really excited about?
In line with the above prediction about fraud detection specifically, we are very excited about ‘regtech’. FC enables the development of innovative solutions to address fraud, systemic risk, and back office operations – all with scalable tech that addresses the privacy concerns that often arise with AI. Back offices lacking resources and regulators will appreciate the low-cost software alternative to manual processing. Participants in the financial system will be pleased because of the increased speed and reduced risk. And privacy watchdogs will be pleased because of the focus on preserving privacy. Rhino has had success deploying these solutions among healthcare organizations and their regulatory bodies, and we are excited by the traction we’re now seeing in financial services.