Federated Learning & Confidential Computing
What is Federated Learning?
As we mentioned in our previous Blog Post, one of the most interesting applications of Confidential Computing is in the rapidly developing field of Federated Learning. Given this topic is more complex, we felt it deserved its own special attention. So what exactly is Federated Learning?
It is also known as distributed machine learning. We all know that the larger the data set - the more accurate the results. However, in today’s world data is quite literally digital gold. The global big data and business analytics market was valued at 169 billion$ in 2018 and is expected to grow to 274 billion$ in 2022. So naturally the dilemma that is faced is that everyone wants access to improved data analysis results but is not willing to share their data. Additionally, most institutions face stringent data protection regulations which further restrict their ability to share data - even if they wanted to!
Federated Learning allows for machine learning algorithms to access a wide range of data sets situated at different locations. It enables companies (such as banks, insurers, manufactures etc.) to collaborate on the development of data models and access improved results, but without needing to directly share secure data with each other. This means that the algorithms gain exposure to far more data than each individual company has at its disposal. Federated Learning decentralizes machine learning by removing the need to share data or centralise it in a single location.
So, we can all agree that Federated Learning is great. It is a solution which avoids data sharing but rather focuses on model sharing. However, there is still the problem with the fact that most providers of Algorithms are recultant to share their own Intellectual Property with those 3rd parties. This is understandable as they have built their business around their machine learning models and algorithims. This together with the risk the data is at when it is ‘in-use’ presents a significant problem to this otherwise elegant solution. This is where Confidential Computing comes in.
How does Confidential Computing enable Federated Learning?
As we have indicated - Confidential Computing protects data during processing, and when combined with storage and network encryption with exclusive control of encryption keys, provides end-to-end data security in the cloud or on-premise. It isolates sensitive data in a protected CPU enclave during processing. The contents of the enclave - the data being processed, and the techniques used to process it - are accessible only to authorized programming code, and invisible and unknowable to anything or anyone else.
The use of these Secure Enclaves, or Trusted Execution Environments (TEE’s) solves the final problem in Federated Learning by protecting both the Data and the proprietary algorithms. If malware or other unauthorized code attempts to access the keys - or if the authorized code is hacked or altered in any way - the TEE denies access to the keys and cancels the computation. This is exactly what is offered through MADANA Core.
Are you interested in MADANA’s confidential computing solution? Then please email us at [email protected]!
