Research

Research Interests

The main three themes of my research career are:

1. Applied Cryptography
2. Machine/Deep Learning
3. Hardware Acceleration

Specific research interests include but are not limited to:

1. Parallel Processing and High-Performance Computing
2. Privacy-Preserving Technologies
3. Trustworthy Machine Learning
4. Machine Learning for Cybersecurity
5. Digital Signal Processing
6. Computer Systems Performance Modelling (Simulation, Queueing Networks and many others)
7. Combinatorial Optimization
8. Evolutionary Algorithms

Grants and Contract Awards

Feb 2020 – Aug 2021
RIE2020 Advanced Manufacturing and Engineering (AME) Programmtic Programme (Award A19E3b0099). 9.9 Million SGD.
Principle Investigator for a multi-year, multi-million dollar research and development effort funded by the Singaporean government to optimize fully homomorphic encryption (HE) for advanced manufacturing applications such as Design of Experiments. I was responsible of Work Package II – Strengthening HE fundamentals tasked with:

1. designing fast algorithms for primitive HE computations,
2. extending HE arithmetic set, and
3. increasing the usability of HE to ease application development.
   10 peer-reviewed conference and journal articles, 2 patents, 3 software IPs, and 1 software library were generated from this project.

Research Products

1. GPU acceleration of different fully homomorphic encryption schemes such as BFV and CKKS.
2. Applying fully homomorphic encryption in creating privacy-preserving machine and deep learning using encrypted data.
3. Homomorphic CNNs: developing a privacy-preserving image classifier that performs inference (prediction) of CNNs on encrypted images from MNIST, CIFAR-10 and diabetic retinopathy medical images. Implemented on CPU (using SEAL) and GPU (using A*FHE).
4. PrivFT (private fasttext): developing a privacy-preserving NLP text classifier based on the bag-of-words model using FHE. PrivFT can perform both training and prediction on encrypted text. Implemented on CPU (using SEAL) and GPU (using A*FHE).
5. GI-FHE: developing privacy-preserving genotype imputation models that impute the missing genotypes of encrypted DNA sequences using FHE. Implemented on CPU (using SEAL).
6. AML-FHE: developing a privacy-preserving anti-money laundry system that predicts illicit encrypted transactions using FHE. Implemented on CPU (using SEAL).
7. CASHE: developing a privacy-preserving spacecraft conjunction analysis tool using FHE. CASHE can predict the collision probability of flying objects using encrypted orbital information. Implemented on CPU (using SEAL).
8. Multi-processor task scheduler via Particle Swarm Optimization.

Current Research Activities

Apr 2021 – Jun 2024
DPRIVE - Data Protection in Virtual Environments – 15 Million USD. TREBUCHET Team
Associate Principle Investigator for a multi-year, multi-million dollar DARPA-funded research and development effort to build a HE chip for homomorphic machine and deep learning applications via ASIC technology. I am mainly responsible for:

1. designing novel cryptographic capabilities of HE to support machine and deep learning applications such as logistic regression training, CNN inference and training,
2. developing initial prototypes of the said cryptographic capabilities
3. coordination and knowledge transferal between the software and hardware subgroups in the TREBUCHET team, and
4. developing workload charecterization and performance models on various harware acceleration technogies.