Research Areas
Our work spans multiple disciplines at the intersection of cryptography, machine learning, and systems engineering.
FHE Circuit Optimization
Developing novel techniques to reduce computational overhead in homomorphic operations while maintaining security guarantees.
Neural Architecture Design
Creating ML architectures specifically designed for encrypted computation, balancing accuracy with FHE compatibility.
Cryptographic Protocols
Advancing multi-party computation protocols and hybrid encryption schemes for practical privacy-preserving systems.
Hardware Acceleration
Building specialized hardware and GPU kernels optimized for homomorphic encryption operations.
Quantization Methods
Researching low-precision arithmetic and adaptive quantization for efficient encrypted inference.
Privacy-Utility Tradeoffs
Formalizing and optimizing the balance between privacy guarantees, model accuracy, and computational efficiency.
Research in Progress
We're actively working on several research papers exploring the frontiers of privacy-preserving computation.
Efficient Homomorphic Training of Deep Neural Networks (In Preparation)
Velum Labs Team
Novel gradient approximation methods that enable practical backpropagation on encrypted data.
Clifford Neural Networks: Geometric Structures for Privacy-Preserving ML (In Preparation)
Velum Labs Team
A new class of neural networks built on Clifford algebras that maintain geometric equivariance under encryption.
Hardware-Accelerated FHE: A Systematic Study (In Preparation)
Velum Labs Team
Analysis of GPU and FPGA acceleration techniques for homomorphic operations.