An Investigation of Pseudonymization
Techniques in Decentralized Transactions 1Wonkwang
University, Jeonbuk,
Iksan City 54538, Republic of Korea 2Pukyong National University, Busan 48513, Republic of Korea mfirdaus@pukyong.ac.kr, khrhee@pknu.ac.kr
Abstract Decentralized learning (DL) enables several devices
to assemble deep learning models while keeping their private training data on
the device. Rather than uploading the training data and model to the server,
cross-silo DL only sends the local gradients gradually to the aggregation
server back and forth. Hence, DL can provide privacy training of machine
learning. Nevertheless, cross-silo DL lacks the proper incentive mechanism
for the clients. Thanks to the blockchain, smart contracts (SCs) can address
the concerns by providing immutable data records which are self-executing and
tamper-proof to failures. Yet, the records of blockchain transactions are publicly
visible, which can leak valuable clients' information as analytical systems
become more sophisticated. We leverage the Monero
(XMR) protocols to be adjusted into cross-silo DL transactions over wireless
networks to address the issues. Concurrently, we investigate the performance
of constructed protocols embedded into blockchain smart contracts. This paper
also reports and analyzes an empirical investigation of several privacy
preservation techniques in decentralized transactions. Overall, the performance
results satisfy the design goals. Our observations fill the current
literature gap concerning an up-to-date systematic mapping study, not to
mention extensive techniques in preserving privacy for cross-silo DL combined
with blockchain. Keywords: blockchain-based incentive,
decentralized learning, pseudonymization protocols, smart contract +: Corresponding author: Kyung-Hyune
Rhee
Journal of Internet Services and
Information Security (JISIS), 11(4): 1-18, November 2021 |