CAREER: Practical Verifiable Computation on Private Data

Modern computing increasingly relies on cloud services and decentralized networks, where users often outsource computation to servers they cannot control or fully trust. Artificial intelligence systems are increasingly deployed in domains that rely on sensitive data, such as healthcare and finance, raising concerns about the privacy of training data, the security of models, and trust in their outputs. Models often depend on sensitive data and perform inference on private inputs, especially in healthcare and biomedical research where legal and ethical constraints limit data sharing. These developments raise two fundamental challenges: ensuring privacy of inputs and ensuring integrity of outputs. Cryptographic techniques address both. Fully Homomorphic Encryption (FHE) allows a server to compute directly on encrypted data so that sensitive inputs remain hidden, while cryptographic proof systems allow a server to generate a short proof that a computation was performed correctly, which a client can verify efficiently even when the computation itself is expensive. Together, these tools enable verifiable computation on private data, in which the server learns nothing about the data or the result, yet the client can efficiently verify that the output was computed correctly. However, despite major advances in both FHE and cryptographic proof systems, existing approaches for verifiable computation on encrypted data remain too inefficient for practical deployment. This project develops new cryptographic techniques and systems to make verifiable computation on private data practical. The project also integrates research and education through university courses and outreach activities introducing students to modern privacy and security technologies. This project pursues three interrelated research directions. First, it develops faster cryptographic proof systems with smaller proofs and efficient verification that remain efficient across the different numerical representations used by fully homomorphic encryption systems for both plaintext and encrypted data. Second, it constructs end-to-end systems for verifiable FHE, including approaches in which cryptographic proofs attest to the correctness of operations on encrypted data and approaches in which proofs are generated as part of the homomorphic computation. Third, it develops interactive protocols for verifiable delegation of computation that go beyond FHE and are tailored to tasks such as matrix multiplication and neural network inference. Across these directions, the project will produce new constructions, prototype implementations, and performance evaluations that bridge the gap between modern cryptographic theory and practical systems for secure and trustworthy computation on sensitive data. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2544307 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT,01002930DB NSF RESEARCH & RELATED ACTIVIT,01003031DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Benjamin Fisch | Institution: Yale University, NEW HAVEN, CT | Award Amount: $297,954 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2544307 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2544307.html