Abdoul Ahad FALL

Context: Operational security for transactions via secure messaging. Research on a secure, resilient, and post-quantum-compatible federated learning system with adaptive interactions.

• Responsibilities:
  
  Supervision of cryptographic object management: Ensure the generation, lifecycle tracking, inventory, and auditability of cryptographic objects in compliance with applicable standards.
• Management of sensitive information media (SIC): Secure, administer, and control all media containing critical information.
  
• HSM and PKI infrastructure management: Oversee, maintain, and optimize Hardware Security Modules (HSM) essential for key protection.
  
• Cryptographic evolution steering: Participate in strategic discussions on cryptographic practices and propose improvements to existing processes.
  
• Cryptographic project leadership: Manage end-to-end projects, from planning to implementation, coordinating stakeholders while meeting deadlines and quality objectives.
  
• Automation and deployment of secret management tools: Design and implement automated solutions for efficient cryptographic secret management.

Actions:

• Studied a hybrid post-quantum security scheme using CRYSTALS-Kyber for key exchange and CRYSTALS-Dilithium or SPHINCS+ for signing and verifying the integrity of federated learning model updates.
• Integrated Zero-Knowledge Proofs (ZKP) to prevent inference attacks on transmitted model parameters.
• Developed a post-quantum secure aggregation protocol with interactive update validation, ensuring the authenticity of each contribution while preserving the confidentiality of model gradients.
• Strengthened resilience to adversarial and quantum attacks by leveraging explainable AI algorithms capable of detecting malicious or poisoned updates (considered as an alternative perspective).
• Optimized the system for distributed environments (cloud, edge computing, IoT), focusing on low cryptographic footprint and full interoperability with existing infrastructures.
• Migrated to post-quantum cryptography through implementation of post-quantum and hybrid protocols.

Tools and open-source libraries:
Open Quantum Safe, Liboqs
TLS, KEMTLS
C language

Context:
Research and development on Fully Homomorphic Encryption over the Torus, including a proof of concept for hardware implementation.

Actions:

• State-of-the-art review of fully homomorphic encryption schemes: BGV, BFV, GSW, TFHE, CKKS, CHIMERA.
• Study of hardware implementation in an IoT context using CubeIDE STM32 (C language).
• Analysis of lattice problems: Shortest Vector Problem (SVP), Closest Vector Problem (CVP), Unique Vector Extension (UVE).
• Security proofs: study of IND-CPA and IND-CPA' (chosen plaintext attacks).
• Feasibility assessment: testing and decision-making regarding hardware implementation.
• Study of side-channel attacks: analysis of entropy related to decryption time and number of encrypted bits, timing attacks, and mathematical modeling of entropy.
• Neural network layer analysis:
  Identification of commonly used layers.
  Study of layers that can be homomorphically evaluated via leveled operations.
  Study of non-linear or more complex layers requiring one or more bootstrappings (PBS).

Tools:
OpenSSL
Languages: Python, C

Machine Learning: Exploration of various learning paradigms.

Models:

• Regression
• Classification

Model Structure:

• Implementation using Python

Optimization Algorithms:

• Linear Regression
• Logistic Regression
• Neural Networks

Model Updates:

• Gradient Descent

Implementation and Training:

• Development of a model based on linear regression