zero-order-federated-learning-he
Privacy-enhanced zero-order federated learning using multi-key homomorphic encryption (xMK-CKKS) over wireless channels. Methodology for secure FL aggregation without channel estimation, supporting N-1 client compromise tolerance. Use when designing privacy-preserving federated learning, multi-key HE protocols, or wireless FL systems.
소스 정보
- 저장소
- hiyenwong/ai_collection
- 최근 소스 활동
- 2026년 6월 4일 13:32
- 감지된 SKILL.md 언어
- 영어
- 스타
- 2
- 포크
- 0
설치 방법
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
소스 파일 검토
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
SKILL.md 표시 중
SKILL.md
소스 지침 · 읽기 전용 미리보기- name
- zero-order-federated-learning-he
- description
- Privacy-enhanced zero-order federated learning using multi-key homomorphic encryption (xMK-CKKS) over wireless channels. Methodology for secure FL aggregation without channel estimation, supporting N-1 client compromise tolerance. Use when designing privacy-preserving federated learning, multi-key HE protocols, or wireless FL systems.
- license
- Complete terms in LICENSE.txt
- metadata
- {"arxiv_id":"2605.30123","published":"2026-05-28","authors":"Anthony Ayli, Khalil Harris, Jihad Fahs, Mohamad Assaad","tags":["federated-learning","homomorphic-encryption","xmk-ckks","privacy","wireless-ml","zero-order"]}
# Zero-Order Federated Learning with Multi-Key HE
## Overview
Privacy-preserving federated learning protocol combining zero-order FL (one encrypted scalar per device per round) with multi-key homomorphic encryption (xMK-CKKS) over wireless channels — **without channel estimation or pre-equalization**.
## Core Architecture
### Protocol Design (Four-Phase)
1. **Partial Public Key Retransmission**: Devices transmit partial public keys through the same channel realization
2. **Ciphertext Retransmission**: Partial ciphertexts retransmitted through same channel
3. **Algebraic Cancellation**: Dominant large-modulus encryption terms cancel during decryption (same channel → same fading)
4. **Aggregation**: Server aggregates encrypted scalars, decryption noise preserves O(1/√K) convergence rate
### Key Innovation
- **No channel estimation required**: Same-channel retransmission enables algebraic cancellation of fading terms
- **Client-level security**: Each device has its own secret key (multi-key HE vs single-key)
- **Compromise tolerance**: Secure against honest-but-curious server colluding with up to N-1 clients
- **Communication efficiency**: Overhead independent of model dimension (one scalar per round)
### Convergence Analysis
Decoded encryption noise preserves O(1/√K) convergence rate up to negligible noise floor, validated on MNIST with slowly varying LoS-dominant channels.
## When to Use
- Privacy-preserving FL with untrusted clients
- Wireless FL with channel fading (LoS-dominant, slowly varying)
- Scenarios where single-key HE is insufficient (multi-party trust)
- Bandwidth-constrained FL (zero-order scalar transmission)
## Pitfalls
- Requires slowly varying channel (retransmission assumes same channel realization)
- LoS-dominant channels work best; rich scattering may break cancellation
- Multi-key HE has higher computation overhead than single-key
---
*Reference: Ayli et al. (2026) "Privacy-Enhanced Zero-Order Federated Learning via xMK-CKKS over Wireless Channels" arXiv:2605.30123*
GitHub에서 보기