ML Security PapersEByFTVeS: Efficient Byzantine Fault Tolerant-based Verifiable Secret-sharing in Distributed Privacy-preserving Machine Learning
Zhen Li 1, Zijian Zhang 1, Wenjin Yang 1, Pengbo Wang 1, Zhaoqi Wang 1, Yan Wu 1, Xuyang Liu 1,2, Jing Sun 2
Published on arXiv
2509.12899
Data Poisoning Attack
OWASP ML Top 10 — ML02
Key Finding
EByFTVeS provably eliminates the ACuMPA attack vector and achieves superior computational efficiency compared to state-of-the-art VSS-based DPML schemes in empirical evaluations.
EByFTVeS / ACuMPA
Novel technique introduced
Verifiable Secret Sharing (VSS) has been widespread in Distributed Privacy-preserving Machine Learning (DPML), because invalid shares from malicious dealers or participants can be recognized by verifying the commitment of the received shares for honest participants. However, the consistency and the computation and communitation burden of the VSS-based DPML schemes are still two serious challenges. Although Byzantine Fault Tolerance (BFT) system has been brought to guarantee the consistency and improve the efficiency of the existing VSS-based DPML schemes recently, we explore an Adaptive Share Delay Provision (ASDP) strategy, and launch an ASDP-based Customized Model Poisoning Attack (ACuMPA) for certain participants in this paper. We theoretically analyzed why the ASDP strategy and the ACuMPA algorithm works to the existing schemes. Next, we propose an [E]fficient [By]zantine [F]ault [T]olerant-based [Ve]rifiable [S]ecret-sharing (EByFTVeS) scheme. Finally, the validity, liveness, consistency and privacy of the EByFTVeS scheme are theoretically analyzed, while the efficiency of the EByFTVeS scheme outperforms that of the-state-of-art VSS scheme according to comparative experiment results.
Key Contributions
- Formally defines the Adaptive Share Delay Provision (ASDP) timing-based vulnerability in BFT-integrated VSS schemes, proving fabricated shares are accepted by honest participants
- Introduces ACuMPA, a customized model poisoning attack that exploits ASDP to inject tailored malicious gradients that evade cosine-similarity filtering defenses
- Proposes EByFTVeS, a BFT consensus-driven VSS scheme that enforces strict temporal locking of share commitments, eliminating the timing window required for ACuMPA while outperforming state-of-the-art VSS in computational efficiency
🛡️ Threat Analysis
ACuMPA is a Byzantine poisoning attack where a malicious dealer exploits timing asymmetry (ASDP) to inject crafted gradient shares during distributed training, corrupting global model aggregation while bypassing cosine-similarity defenses. EByFTVeS defends via BFT consensus-enforced synchronization of share commitments — a robust aggregation defense against malicious participants in distributed/federated ML training.