About

Third-party data are on their way out, with Google and Apple pioneering this trend. Most technology platforms are killing tracking cookies and imposing cumbersome privacy labels. This means that first-party data are becoming critical for mobile app publishers to understand their users. However, even mining first party data comes with a lot of privacy concerns from users, as the Whatsapp case demonstrates, and regulators (GDPR, CCPA). Until now, it was impossible to combine personalization and privacy - but a new technology called Federated Learning is now here to change this. We are the first to enable mobile marketers tailor experiences on an individual level, while guaranteeing privacy. This is a billion dollar market, just taking into consideration North America. Our long-term vision is to build the infrastructure for private machine learning across the billions of edge devices, which will reshape the way the world extracts knowledge.

Solution & Product

Lerna is building a novel solution that helps app publishers personalize their offering. The solution comprises two components: a mobile library and a cloud-based service. We categorize our customers’ users based on their context and optimal engagement timing. Subsequently, we automate relevant actions, e.g. our cloud service exposes APIs and Webhooks that send push notifications with optimal content at the most relevant moment. We can offer more accurate predictions, as the amount of data available on the device is richer than what is used by vendors currently. In comparison to other solutions, we can offer more accurate predictions, as the amount of data available on the device is richer than that is used by vendors currently. By tapping into the sensor data as an additional source, we can infer the user’s context. The Lerna product is built on four pillars: 1.tIt unlocks out-of-the box mining of data directly on mobile devices 2.tIt increases the accuracy of this predictive learning, as its device libraries tap into an unprecedentedly rich dataset that never leaves the devices 3.tIt builds the world’s most efficient privacy and security layer on top of mobile federated learning 4.tIt is simple for developers to integrate Lerna as a black-box without any knowledge of AI, privacy or federated learning

Innovation & Scalability

For Federated Learning to be truly privacy-preserving, there is a layer of privacy and security tools that need to be stacked up on top of the underlying learning infrastructure. However, in today’s FL frameworks, this privacy and security layer acts as a bottleneck and does not scale smoothly. We solve this by introducing a private trusted 3rd party, which ensures that we can achieve: ●t50x faster FL than current setups ●tHigher accuracy via private transfer learning For this purpose we use cutting-edge technologies, such as Secure Multi-Party Computation and Differential Privacy. There are massive growth opportunities in extending our solution. We are planning to turn Lerna into the underlying platform and stamp of approval for private Machine Learning across the billions of edge devices in the world. This goes far beyond mobile, and includes wearables, driverless cars and manufacturing IoT.

Traction & achievements

We have already built our MVP, enabling Android developers optimize the timing of their campaign outreach. We have closed our pre-seed round, with participation from some of the best institutional investors in Greece.

Team & Why Us

Our team possesses a deep domain expertise to solve the problem. 1.tUnderstanding of the privacy market: The two co-founders previously founded a startup automating compliance and de-identification for data mining. 2.tInvolvement in open source communities: we have both contributed to working groups in the most successful open source community for federated learning. 3.tExpertise in ML products and private data mining: Georgios Depastas has led ML products at Barclays, where he served as the Head of the Real-time Big Data Analytics portfolio. Georgios Kellaris is an expert in data privacy and secure data mining; he worked as a postdoctoral researcher in the domain jointly at Harvard and Boston University.

Funding Needs

We concluded our pre-seed round earlier this year, hence we are not currently raising.