Advances Post-Quantum Aggregation with Code-Based Homomorphic Encryption and LPN

Why It Matters

The protocol offers a quantum‑resistant alternative to lattice‑based aggregation, strengthening privacy‑preserving analytics for federated learning and multi‑party computation. Its efficiency and provable security accelerate real‑world adoption of post‑quantum cryptography.

Key Takeaways

• Introduces code‑based LPN homomorphic encryption for aggregation.
• Committee‑based decryptor uses secret sharing to protect inputs.
• CRT optimisation cuts communication overhead in LPN schemes.
• Hint‑LPN proven equivalent to standard LPN security.
• Outperforms info‑theoretic protocols in certain parameter regimes.

Pulse Analysis

Secure aggregation is a cornerstone of privacy‑preserving machine learning, yet the looming advent of quantum computers threatens many existing lattice‑based constructions. By turning to code‑based cryptography and the Learning Parity with Noise problem, the Munich team taps a hardness assumption believed to resist quantum attacks, expanding the toolbox for cryptographic agility. This shift not only diversifies post‑quantum options but also aligns with industry calls for alternatives that can withstand future cryptanalytic breakthroughs.

The paper’s technical contributions are threefold. First, it introduces a key‑ and message‑additive homomorphic encryption scheme that enables direct summation of encrypted inputs. Second, a committee‑based decryptor, implemented through secret sharing, ensures that no single party can uncover individual contributions, bolstering trust in multi‑party settings. Third, a Chinese Remainder Theorem optimisation decomposes the aggregation across smaller moduli, dramatically slashing the bandwidth typically required by LPN‑based protocols. Security is rigorously proved under a novel Hint‑LPN assumption, which the authors demonstrate to be equivalent to standard LPN for carefully chosen parameters, offering a solid foundation for future code‑based designs.

Performance measurements reveal that, while LPN‑based schemes can incur higher raw communication than lattice alternatives, the CRT‑driven reductions enable the new protocol to outperform information‑theoretically secure methods in targeted parameter regimes. The accompanying SageMath tool streamlines parameter selection, lowering the barrier for practitioners to adopt the scheme in federated learning, collaborative analytics, and other data‑intensive domains. As enterprises grapple with regulatory privacy mandates and the inevitability of quantum threats, such efficient, provably secure aggregation mechanisms are poised to become integral components of next‑generation data pipelines.