Startup holds 'secret' to analyzing private data without accessing it
Jakob Odersky, Dimitar Jetchev, Manohar Jonnalagedda and Iraklis Leontiadis of Inpher, at the EPFL Scala Days conference in June. © Dimitar Jetchev
EPFL Innovation Park startup Inpher has developed a computing platform that maximizes security by allowing users to see only the results of analytics performed on distributed, private data – and not the data itself. Collaborators hired from the School of Computer and Communication Sciences (IC), who have brought their expertise in cryptography, security, compilers and programming languages to Inpher, have been key to the company’s success.
Inpher’s Secret Computing® cryptographic platform, based on a technology called secure multiparty computation, has applications in finance, health, manufacturing and beyond. The software has particular potential for companies in countries like Switzerland, where laws prohibit the sharing of customer data across borders.
“This technology guarantees that you are not learning anything about input data, but only about the function being performed on the data. Therefore, it falls outside scope of the [European Union General Data Protection Regulation] GDPR,” explains Dimitar Jetchev, Inpher co-founder and Chief Technology Officer.
From fintech to flight
Inpher’s secret computing platform allows artificial intelligence models to draw conclusions from encrypted data from multiple distributed private data sources without compromising security.
To explain the concept of secure multiparty computation, Jetchev offers an example: “Imagine that you have a secret number, and are in a game with two other players. If you generate three random numbers whose sum is your secret number, and distribute two of the random numbers to the two other players, no one player except you – the data holder – would ever know the secret number itself.”
Such an approach allows machine learning models to be trained on multiple data sources from different regions or countries, resulting in better algorithms without compromising privacy. Insights from the data can also be monetized, again without selling or revealing it.
In the financial sector, which until now has suffered from a lack of aggregated data that can safely be used for analytics, such machine learning models could be used to perform credit risk analyses or detect fraud. Inpher’s technology also has potential for a field called predictive maintenance, which is exactly what it sounds like: the process of estimating how much time remains before a product must be replaced, based on data about how that product is used and what stresses it’s exposed to. Predictive maintenance is especially important in the flight industry for ensuring that all aircraft parts are safe and up to code.
Branching out from Swiss roots
Jetchev, a Swiss National Science Foundation professor in EPFL’s School of Basic Sciences, did his postdoctoral research in the IC’s Laboratory for Cryptographic Algorithms (LACAL). He co-founded Inpher with CEO Jordan Brandt in EPFL’s Innovation Park in 2015.
Today, the company has a New York office, and has been featured in the Wall Street Journal. At the end of last year, the startup reached a major milestone in its development when it received $10 million in Series A financing from the JP Morgan Chase & Co strategic investment fund.
Most of Inpher’s original staff brought cryptography and security expertise from EPFL. Today, there are about 18 employees who have complementary skills in statistics, machine learning and compiler design. Most of the company’s core technology is still developed in at the Innovation Park.
“Inpher’s secret computing platform requires a lot of expertise in compilers and programming languages, and we’ve been very lucky to be near EPFL,” Jetchev says, noting ongoing collaborations with IC’s laboratories for Programming Methods (LAMP), Communications and Applications (LCA), Decentralized and Distributed Systems (DEDIS) and Security and Cryptography (LASEC), as well as the Swiss Data Science Center (SDSC).