Privacy-preserving ML on encrypted data

Jessie Chen, based between Singapore and Cambridge, was selected by Emergent Ventures in its 46th cohort (September 2025) for work on privacy-preserving machine-learning models that operate directly on encrypted data, without exposing the underlying inputs. The award is a discretionary Tyler Cowen grant in the program's typical $1K to $50K range, administered through the Mercatus Center.

The technical area encompasses several related approaches. Fully homomorphic encryption (FHE) lets a server compute on ciphertext and return an encrypted result the client decrypts. Hybrid homomorphic encryption combines FHE with symmetric encryption to reduce ciphertext expansion. Secure multiparty computation lets multiple parties jointly compute a function over inputs none of them reveal. Trusted execution environments are a hardware-attested counterpart. Open research artifacts in the area include HE-MAN (homomorphic inference for ONNX models) and GuardML (hybrid-FHE inference service). The category sits in the identity-trust meta-layer and is adjacent to confidential computing.

The grant target is the kind of setting where regulatory or contractual constraints prevent raw data from leaving its origin, including healthcare, finance, and inter-organization analytics. Beyond the cohort announcement, no further public deliverable has been linked.