Raluca Ada Popa

Digital Library

ACM Grace Murray Hopper Award

USA - 2021

citation

For contributions to the design of more practical distributed systems for secure computation over encrypted data that protect confidentiality against server attacks while maintaining full functionality and low performance overhead

Dr. Raluca Popa has made a substantial impact on the design and building of scalable and practical systems for secure computation and the protection of sensitive data.

Working at the interface between cryptography and systems, she has combined theory with practical implementations, and demonstrated how to protect confidentiality against server attacks while maintaining full functionality and low performance overhead. Instead of just using prohibitively expensive theoretical techniques such as fully homomorphic encryption, Raluca combines specialized cryptographic, privacy-preserving constructs with a smart building-block system design approach that results in practical systems that can run important classes of applications over encrypted data, with low performance overheads.

Popa designed a number of prototype systems, in different application domains, that empirically demonstrate the effectiveness of a whole systems based approach to private computation. In Opaque, DORY, Metal, CryptDB, BlindBox, and Arx, she showed how the utilization of encryption schemes that efficiently support a few carefully identified primitive operations enables performant SQL queries and filtering on encrypted databases and packets. For "cooperative learning," the Helen and Senate prototypes she proposed enables multiple organizations to collaboratively train a machine-learning model or perform data analytics over their combined encrypted data. In Delphi and MUSE, machine learning models execute on the client's input, without revealing the data to the model provider or leaking the model to the client.

Several mainstream commercial databases and systems have adopted the techniques she has pioneered. Her research will likely continue to have a substantial impact in how people think about, design, and build scalable and practical distributed systems for secure computation, storage, and communication.

Press Release

2021 ACM Grace Murray Hopper Award

Raluca Ada Popa, University of California, Berkeley, is the recipient of the 2021 ACM Grace Murray Hopper Award for the design of secure distributed systems. The systems protect confidentiality against attackers with full access to servers while maintaining full functionality.

Popa’s fundamental work of building secure systems focuses on protecting the confidentiality of data stored on remote servers. Cloud computing makes sensitive data more accessible to hackers and insiders, despite the common “faulty” assumption that parts of the server–say the databaseor operating system–are inaccessible and can be “trusted." Popa’s research provides confidentiality guarantees where servers only need to store encrypted data, processing it without decrypting. Thus, hackers see only encrypted data.

Computing on encrypted data, possible in theory, has been prohibitively inefficient in practice. Popa addresses this by replacing generality with building systems for a broad set of applications with common traits, and developing encryption schemes tailored to these application archetypes. In SQL databases, for example, Popa extracts a few primitive operations that support most queries, utilizes encryption schemes that efficiently support these primitives, and thus can perform most computations on encrypted databases.

Popa, as the senior researcher, has designed an astonishing number of prototype systems in different application domains, providing functionality over encrypted data. In Opaque, DORY, Metal, and CryptDB, she showed how the utilization of cryptographic schemes that efficiently support a few carefully identified primitive operations enables performant encrypted databases and file systems. The Helen and Senate prototypes she and her students contributed enable multiple organizations to collaboratively train a machine-learning model or perform data analytics over their combined encrypted data. In Delphi and MUSE, machine learning models execute on the client’s input, without revealing the data to the model provider or leaking the model to the client.