XAIN, the AI startup that specializes in privacy-oriented Federated Machine Learning (FedML), is developing an infrastructure to train artificial intelligence applications through FedML technology, a mechanism that emphasizes data privacy. XAIN’s distributed approach to machine learning, which intends to comply with the European Commission’s General Data Protection Regulations (GDPR), also provides greater efficiency in the way data is trained, marking a major breakthrough in a field otherwise burdened by costly and onerous processes.
When you download facial recognition software onto your phone, your data is usually stored on the central database of the app providing the service. FaceApp, for instance, infuriated the public recently for storing data centrally, though they’re far from the first AI-based app to lack privacy protection measures. Data aggregation is essential for AI technology to work — the question is how to preserve privacy throughout the process. Enter XAIN’s FedML technology.
Instead of training, aggregating, and storing data on one centralized AI model, XAIN’s FedML trains each data set separately. The findings from each separate AI model are communicated and aggregated with the findings of the other data sets, but only the aggregations are stored together. That means each individual data set remains private and there is no need to anonymize or store local data in a central source. This reduces cost and simplifies the data training process without compromising the privacy of local data sets, an attractive feat for businesses.
\XAIN’s FedML technology can be used both internally by companies, and externally in the creation of products. The first application running its training models on XAIN’s FedML technology is ANDY. This solution for automated invoice processing consolidates machine learning knowledge from each customer for whom the app is running, all while keeping the customer’s data private in its respective corporate environment. ANDY is designed for larger enterprises, where the application really utilizes its potential — especially when the trained models are consolidated among various departments within a company, or even between corporate subsidiaries.
“One of the biggest challenges going forward in AI is figuring out how to simplify the training process, which is incredibly onerous, while also keeping data secure,” says Dr. Christian Nagel, Founding Partner at Earlybird Venture Capital and Member of the XAIN Supervisory Board. “FedML addresses that challenge, while also reducing the cost of training.”
Sign up for the free insideBIGDATA newsletter.
Speak Your Mind