Applied Cryptography
Fully homomorphic encryption, lattice-based and post-quantum primitives, secure multi-party computation, and protocol design for real-world deployments.
Research
Research themes, signature work, and ongoing projects in cryptography, privacy-preserving machine learning, and hardware acceleration.
For the agenda behind this work, see my research statement.
For how my work has been used, see my research impact.
Themes
Machine & Deep Learning
Trustworthy and privacy-preserving ML, encrypted inference and training, and machine learning for cybersecurity and intrusion detection.
Hardware Acceleration
GPU and ASIC acceleration of lattice cryptography, parallel processing, high-performance computing, and software–hardware co-design.
Specific Interests
- Fully Homomorphic Encryption
- Privacy-Preserving Machine Learning
- Private LLM Inference
- Encrypted Deep Learning
- Lattice Cryptography
- Post-Quantum Cryptography
- Secure Multi-Party Computation
- Threshold Encryption
- Federated Learning
- Differential Privacy
- Trustworthy Machine Learning
- GPU & ASIC Acceleration
- Software-Hardware Co-design
- High-Performance Computing
- Number-Theoretic Transform
- ML for Cybersecurity
Selected venues
A sample of venues where this work has appeared.
- PNAS Proceedings of the National Academy of Sciences
- NeurIPS Neural Information Processing Systems (PriML)
- IACR TCHES Transactions on Cryptographic Hardware and Embedded Systems
- IACR CiC Communications in Cryptology
- IEEE TPDS Transactions on Parallel and Distributed Systems
- IEEE TETC Transactions on Emerging Topics in Computing
- WAHC Workshop on Encrypted Computing & Applied Homomorphic Cryptography (ACM CCS)
- ISPASS IEEE International Symposium on Performance Analysis of Systems and Software
Signature work
A selection of papers with significant contributions to practical fully homomorphic encryption.
-
OpenFHE: Open-Source FHE Library
Co-author of the leading open-source library for fully homomorphic encryption (BFV, BGV, CKKS, TFHE). Successor to PALISADE, carrying forward a multi-year line of open-source FHE infrastructure. Now standard infrastructure across the FHE research and industry community.
-
Towards the AlexNet Moment for Homomorphic Encryption
HCNN: the first homomorphic convolutional neural network running on encrypted data using GPUs. Demonstrated that encrypted deep learning is practically feasible.
-
RNS Variants of the BFV Homomorphic Encryption Scheme
Introduced the BEHZ and HPS RNS variants of BFV that achieve roughly two orders of magnitude speedup over CPU baselines. Both variants are now standard building blocks in modern FHE libraries.
-
PrivFT: Private and Fast Text Classification with FHE
First demonstration that complex NLP can operate on encrypted text. Performs both training and inference on encrypted documents, enabling privacy-preserving text analytics in sensitive domains.
-
Collaborative Privacy-Preserving Analysis of Oncological Data
Multiparty homomorphic encryption applied to real oncology data, enabling secure cross-institution analysis without exposing patient records. Published in Proceedings of the National Academy of Sciences.
-
High-Performance FV SHE on GPUs (CUDA)
The foundational GPU-FHE result: first single-GPU CUDA implementation of the FV (BFV) scheme. Established the baseline that subsequent multi-GPU and accelerator work builds on.
Current Research
Privacy-Preserving LLM Inference under FHE
Principal Investigator
Making large language model inference run under fully homomorphic encryption with practical latency - bringing private LLM serving within reach of real-world deployments.
- FHE-friendly approximations of transformer building blocks (softmax, GELU, layer norm).
- Bootstrapping schedules and packing strategies tailored to attention and MLP layers.
- End-to-end latency and accuracy evaluation against plaintext baselines.
Past Projects & Grants
DPRIVE: 12 nm FHE Accelerator ASIC (TREBUCHET Team)
Co-Principal Investigator and Technical Lead (since Sept 2024)
A multi-year R&D effort to build a custom 12 nm ASIC for homomorphic machine and deep learning. I directed the final year to successful completion, managing four cross-functional teams (Hardware, Software/ISA, Verification, Applied Crypto) and driving the transition from concept to a fully verified RTL implementation and top-level floorplan.
- 12 nm Process node
- 176 mm² Floorplan
- 1 GHz Timing closure
- 4 Teams led
Includes a custom ISA, microcode scheduler, and node-array architecture. Validated on high-complexity workloads: encrypted CNN training and inference, plus AES transciphering.
Paper: TREBUCHET: Fully Homomorphic Encryption Accelerator for Deep Computation (GOMACTech 2025).
RIE2020 AME Programmatic: Accelerating Homomorphic Encryption
Co-Principal Investigator and Work Package II Lead · Award A19E3b0099
Co-authored the winning proposal for a four-institution Singapore consortium (I2R, NTU, SUTD, NUS) and presented it to A*STAR's review committee. The 3-year programme set out to make FHE deployment-ready for advanced manufacturing, with consulting input from Vinod Vaikuntanathan (MIT), Victor Shoup (NYU), and David Barber (UCL).
- S$9.9M Total programme budget
- 4 Singapore institutions
- 3 International advisors (MIT, NYU, UCL)
- 27 Researchers across the programme
As PI of Work Package II - Strengthening HE Fundamentals, I led the algorithms track: fast algorithms for primitive FHE computations, an extended HE arithmetic set for non-polynomial operations, and tooling to make HE development accessible without deep cryptography expertise.
Programme outputs included 10 peer-reviewed publications, 2 patents, 3 software IPs, and 1 open-source library.