Dustin Ray

Selected Works

• Quantum Trusted Platform Module - A DIY Source of True Quantum Randomness

  Author: Dustin Ray

  Entropy sources in traditional key-management devices suffer from bias in various forms. For example, devices which draw randomness from thermal noise are heavily influenced by ambient temperature. This project aims to use only off-the-shelf components to create a DIY trusted-platform module that sources entropy from the quantum-tunneling effect found when applying a reverse voltage to a Zener diode. Using simple, low cost components, we achieve plausibly-true quantum randomness from this effect. We verify the distribution of noise with the NIST statistical test suite and other techniques.

• Computational Analysis of Plausibly Post-Quantum Secure Recursive Arguments of Knowledge

  Authors: Dustin Ray, Paulo L. Barreto

  Introduces a fully implemented post-quantum secure argument system that compresses unbounded computations into constant-sized proofs. This work extends computational integrity frameworks into the quantum domain, addressing vulnerabilities in cryptographic assumptions against quantum attacks. The system achieves groundbreaking efficiency and scalability, demonstrating real-world performance metrics that surpass classical systems, positioning it as a practical and forward-thinking solution for secure computation.

• Computational Attestations of Polynomial Integrity Towards Verifiable Machine Learning

  Authors: Dustin Ray, Caroline El Jazmi

  Presented at Future Technologies Conference 2024. Demonstrates provable-differential-privacy for large-scale MLaaS, achieving record-breaking verification speeds.

• An End-to-End Framework for Private DGA Detection as a Service

  Authors: Ricardo J. M. Maia, Dustin Ray, Sikha Pentyala, Rafael Dowsley, Martine De Cock, Anderson C. A. Nascimento

  Presents the first comprehensive framework for privacy-preserving classification of domain names into malicious or benign categories using Domain Generation Algorithm (DGA) detection. This innovative solution combines Secure Multi-Party Computation (MPC) and Differential Privacy (DP) to ensure the confidentiality of both DNS traffic and machine learning models. By leveraging post-training quantization techniques, the framework achieves a 23-42% reduction in runtime while maintaining high accuracy, enabling efficient and secure real-world deployment. This work sets a new benchmark for privacy-enhancing technologies in cybersecurity.