Transfer Learning and Federated Learning: A Primer Part Two
by Kristina Drye, on Sep 29, 2021 1:33:31 PM
Advanced technologies abound in the financial industry - but we rarely see an exploration of the different kinds of machine learning, or how they work for compliance. In this two-part blog series, we’ll explore what transfer learning and federated learning are, how they work for compliance practices, and how you can use them together for an ironclad compliance practice.
What Does Transfer Learning Mean for Compliance?
In the first installation of this series, we reviewed that transfer learning is an iterative machine learning technique that uses training data to produce an algorithm, which learns from dataset to dataset. Its precision improves as its lessons accumulate.
Transfer learning has vast potential for the field of compliance. First, algorithms can be trained on any number of problems, or “domains”. These can include, but are not limited to, compliance concerns like human trafficking, corruption, negative news, money laundering, and sanctions offenders.
Second, transfer learning is a method of organizing external data in an efficient and effective way. This means traditional problems facing the compliance space regarding external data- the limits of static lists (they’re not updated as often as we might wish), and the messiness of search engines (they’re unorganized and difficult to navigate) - can be addressed. With transfer learning, the algorithm can be applied to a large dataset, pull information from the internet, and organize it in a meaningful way making the process both more effective and efficient.
Lastly, with each deployment of the algorithm, its knowledge of the problem increases. This is exactly how tools like GOST (Giant Oak Search Technology) work. Risk assessments continue to become more precise, and compliance professionals are able to spend more of their limited time on assessing the entities posing the most risk to an institution.
What Does Federated Learning Mean for Compliance?
As opposed to transfer learning, federated learning is a type of machine learning that trains an algorithm on multiple devices, or sets of data, without actually exchanging the data itself. The key benefit that comes with federated learning is the protection of privacy while sharing data. This is one of the most advanced technologies available to transform a compliance department.
This transformation comes in understanding that a good machine learning tool must continue to process data in order to improve - a precision desperately needed in compliance. At the same time, privacy matters more than ever in the banking industry. How can you train a tool to be its best, while also protecting the privacy of the customers you’re trying to protect?
This is where federated learning can help. By traveling the algorithm rather than the data, data stays put while the tool still learns. With federated learning, your compliance practice can do all that it promises to do: preserve customer’s privacy while protecting them from adverse threats and illicit actors. Tools like Dozer by Consilient are able to achieve this outcome.
How Can You Use Them Together?
Together, federated and transfer learning can be combined for an ironclad compliance practice. Banks have two forms of data to contend with: external data, in the form of the data outside the institution needed to make decisions about its practices, and internal data, in the form of the data protected by the institution and proprietary to it. By employing both transfer learning and federated learning, your compliance department can address both. Transfer learning organizes external data, making screening faster and better; and federated learning trains your algorithms while allowing your data to stay private and protected.
You can find part 1 of Transfer Learning and Federated Learning: A Primer here.