Doctoral Thesis Proposal - Edward Justin Chen
May 28, 2026 2:00PM—3:30PM
Location:
4405
-
Gates and Hillman Centers
Speaker:
EDWARD JUSTIN CHEN,
Ph.D. Student, Computer Science Department, Carnegie Mellon University
https://edwjchen.com/
Advanced analytics and machine learning increasingly depend on data that is sensitive, regulated, and distributed across organizations. Secure multiparty computation (MPC) and fully homomorphic encryption (FHE) offer a path around this barrier by allowing computation over private inputs without revealing the underlying data. Yet despite decades of cryptographic progress, privacy-preserving computation remains difficult to deploy in practice: performance overheads are high, and optimization requires expert domain knowledge.
This thesis argues that compilers are essential for democratizing the use of privacy-preserving computation. Rather than requiring application developers to manually design cryptographic protocols, this work treats the key performance decisions in MPC and FHE as static optimization problems over a program’s intermediate representations (IR). Silph is a compiler that uses integer linear programming (ILP) to optimize the operation and conversion costs of mixed MPC primitives within hybrid MPC protocols. Rotom is an FHE compiler that automatically assigns ciphertext layouts to tensor programs, finding efficient packings that minimize expensive data movement operations. Orbit is another FHE compiler that also uses ILP to jointly place bootstrap and rescale operations by reasoning about the dependencies between ciphertext levels and scales. Together, these systems show that compiler automation can reduce the expertise required to build privacy-preserving applications while improving the performance of the resulting cryptographic programs. The remaining work explores an open problem on how AI-driven methods, such as OpenEvolve, operating over compiler IRs and cost models can discover better cryptographic optimization strategies.
Thesis Committee:
Wenting Zheng (Co-Chair)
Fraser Brown (Co-Chair)
Bryan Parno
Alex Ozdemir (Georgia Institute of Technology)
Srini Devadas (Massachusetts Institute of Technology)
Contact
Matt Stewart