Content uploaded by Muhammad Firdaus
Author content
All content in this area was uploaded by Muhammad Firdaus on Aug 04, 2022
Content may be subject to copyright.
An Investigation of Pseudonymization Techniques in
Decentralized Transactions
Sandi Rahmadika1, Muhammad Firdaus2, Yong-Hwan Lee1, and Kyung-Hyune Rhee2*
1Wonkwang University, Jeonbuk, Iksan City 54538, Republic of Korea
ndiikaa@gmail.com, hwany1458@empas.com
2Pukyong National University, Busan 48513, Republic of Korea
mfirdaus@pukyong.ac.kr, khrhee@pknu.ac.kr
Received: September 3, 2021; Accepted: November 2, 2021; Published: November 30, 2021
Abstract
Decentralized learning (DL) enables several devices to assemble deep learning models while keep-
ing 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 re-
sults 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
1 Introduction
The vigorous utilization of internet-based information systems that rely on a decentralized approach
has been broadly researched by academia, developers, and industries. The foremost objective of the
decentralized approach is to address the communication bottleneck issues and memory usage of the con-
ventional centralized system [22]. The paradigm of a robust concentrated approach (relies on a single
node) via wireless networks has been gradually shifting toward decentralized practices, such as financial
services, healthcare records, any forms of digital rights, and intellectual property. Blockchain technology
through Bitcoin cryptocurrency and decentralized learning are the most prominent practical adoptions of
decentralized approaches. Concurrently, decentralized ledger and artificial intelligence (AI) are expedi-
tiously converging to address many critical challenges. Blockchain technology arguably is an ingenious
invention, the brainchild of a user or group known by Satoshi Nakamoto’s pseudonym [11] that make
Journal of Internet Services and Information Security (JISIS), volume: 11, number: 4 (November), pp. 1-18
DOI:10.22667/JISIS.2021.11.30.001
*Corresponding author: Department of IT Convergence and Application Engineering, Pukyong National University,
Yongso-ro 45, Nam-gu, Busan (48513), Republic of Korea. Telp: +82-(0)51-6296247, Fax: +82-(0)51-6264887
1
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
smart contracts are all the rage in the blockchain world these days. Many pundits claim SCs will convey
an entirely new paradigm that forever changes how the parties write contracts and conduct business. The
successfully conducted blockchain transactions are candidly available in the network and can be accessed
through the user interface by blockchain entities. Thus, it has been practiced in various disciplines of
science.
Linked to the Bitcoin and Ethereum blockchain, distributed learning also relies on the decentralized
approach to collectively building the deep learning model from multiple devices. In contrast to conven-
tional machine learning, where the clients process the training model centrally, FL allows the clients
to build the artificial intelligence (AI) model by sending the updated gradient values to the aggregation
server without revealing the dataset [7]. Accordingly, private data remain confidential (the DL preserves
privacy for clients by design). The DL-based schemes, such as federated learning, lack the proper in-
centive mechanism to motivate clients to improve AI models. Several applications do not even provide
a reward for clients. Blockchain with SC features can be a solution to tackle the incentive mechanism
issues. Nevertheless, directly employing SCs threatens the clients’ privacy since SCs are transparent and
readily available in the blockchain user interface. The existing smart contracts-based solution, such as
Ethereum smart contract and Hyperledger chain-code, only perform a simple computation that cannot
satisfy the application of real-world AI. The developers can program the codes as a self-execute program
without third-party involvement. AI with smart contract integration may render a more resilient and effi-
cient path for a decentralized interactive system for the parties. In short, SCs in DL must be thoroughly
investigated when these technologies are implemented in a system with profoundly confidential infor-
mation, such as federated identity (federated electronic identity) [5] and digital forensics [32]. Precisely,
complementary protocols need to be supplemented.
Privacy-awareness in the smart contract and decentralized learning is part of a flaw that must be
considered in wireless network environments. Melis et al. [30] surprisingly claimed that the observer
could infer the presence of exact data points of clients’ datasets with particular assumptions. While
SCs present client’s transactions visible to the observer. The record of transactions can be accessed
anytime, and the value of data is noticeable. Transparency is one of the concrete features of the SC
blockchain. However, this feature is not desirable for various cases, especially in decentralized learning
with highly confidential data such as medical records, biometric data, employee data, sexual orientation,
philosophical beliefs, and so forth. For these reasons, the relationship between the data used in training
and the owner needs to be obscured. This research explores the techniques of privacy preservation
for cross-silo decentralized learning in untrusted wireless networks. We investigate the application of
decentralized transactions such as decentralized learning by referring to the federated learning principles
(developed by the Google AI team) [16] as a case study for an incentive scheme running on blockchain
SCs. We also analyze the extant protocols in enabling reliable and intelligent system orchestration for 5G
networks and beyond running on a mobile edge computing architecture, AI, and blockchain technology.
An untraceable incentive mechanism based on the data used by leveraging Ethereum smart contracts is
also detailed. A proportional incentive scheme can trigger the entities to contribute to maintaining cross-
silo DL transactions continuously. All processes within decentralized transactions are unlinkable to the
parties. In short, the transactions occur without revealing parties’ information values, which is a part of
privacy-awareness in decentralized approaches.
The rest of this paper is arranged as follows. Section 2 presents an in-depth overview of distributed
ledger technology with a public ledger embodying transaction records along with blockchain application
in the cross-silo decentralized learning (the new technology of trust). The motivation of the paper is
outlined in Section 3. Whilst Section 4 elaborates on the privacy awareness in decentralized approaches
and the linkability concerns in the smart contracts as a revenue mechanism. We also describe the potential
vulnerabilities and defenses in this section. A concrete scheme of secure decentralized transactions is
detailed in Section 5. Finally, we draw some concluding remarks in Section 6.
2
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
2 Blockchain Insights and Cross-Silo Decentralized Learning
This section provides the essential background related to the decentralized ledger, blockchain tables, and
decentralized learning. These approaches are pioneers of the new technology of trust where decentral-
ization is a fundamental component. The implementation of blockchain in decentralized collaborative
learning is also detailed.
2.1 The New Technology of Trust
Blockchain smart contract removes trade or service agreement from the realm of static documents that
require human management. Smart contracts transform into automation tools that manage complex
transactions in the decentralized system. In 2015, Ethereum appeared to the public which adding Turing-
Complete smart contracts to the blockchain. Ethereum performs more complex computations, and it
manages more responses compared to Bitcoin. However, it is not a self-evolving code. Ethereum is
a collection of purely rule-based and recursive programs. A recent study conducted by Kiffer et al.
[20] showed that the smart contract’s diversity is direct copies of other contracts. Moreover, the smart
contracts ecosystem has a considerable lack of diversity since the code is used extensively. Nevertheless,
the concern is gradually tackled since various approaches have risen, including in blending quantum and
cloud computing [8], artificial intelligence (AI), and blockchain smart contracts.
Figure 1: Append-only blockchain tables in general (tamper-resistant property) [1].
Blockchain tables are used to deploy centralized blockchain applications with a central authority,
namely the Oracle database. This centralized form provides an organization with more customizability
and authority to decide who can participate in the network. The authorized users are different database
users who trust the Oracle database to manage tamper-proof blockchain transactions. Compared to fully
decentralized blockchains, this approach is helpful in situations where a higher throughput and lower
latency are favoured over consensus selection. Blockchain tables, in general, provide application trans-
parency security from frauds by other users in the blockchain peer-to-peer network. The frauds can be
recognized by verifying rows in the Oracle table. In this sense, this process re-performs the hash value
and confirms it with the corresponding value stored.
2.2 Blockchain in Cross-Silo DL
Decentralized learning combines decentralized computation and AI learning models to enhance model
training and minimize the risk of privacy breaches in conventional AI techniques. In this context, the
3
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
conventional AI technique, such as machine learning, suffers from severe privacy leakage risk by central-
izing and aggregating the client’s training data that contain private information on a centralized server.
Thus, DL as a decentralized machine learning paradigm allows the clients to collaboratively perform
AI training without giving raw data containing the client’s private information to the central aggregator
[18]. The DL approach empowers the clients to perform a local training model that never leaves their
own devices. In this sense, the client’s raw data is only used to train and update a current global model
and send an updated model to the central aggregator in each iteration. Then, the central aggregator gener-
ates a new global model by aggregating these updated and trained models gathered from the participated
clients to be used in the next iteration. This process is repeated in multiple iterations until the global
model achieves a particular accuracy [2].
Based on application characteristics and client setting, DL is divided into two types, i.e., cross-
device DL and cross-silo DL. Cross-device DL admits a massive number of clients to participate in
model training. Currently, this setting has been widely deployed in consumer digital products, such as
Gboard mobile keyboards [15], Android Message [45], and Pixel phones. Contrary, in the context of
client setting, cross-silo DL relatively provides only a small number of clients by promising better client
reliability and data availability than cross-device DL. In this sense, cross-silo DL ensures that all clients
are almost always available and relatively few failures. Further, cross-silo DL can be relevant for sharing
incentives among clients who train the model using their data for system improvement.
Although cross-silo DL brings several advantages, the existing framework still potentially experi-
ences various adversarial attacks and concerns, including malicious clients and false data, a single point
of failure (SPoF) issue, and the lack of incentives. In order to address these challenges, several works
have been proposed the notion of merging the merits of blockchain and DL for the next-generation wire-
less network [41]. The blockchain-assisted DL approach is used to enhance client privacy and security
by recording transactions in immutable distributed ledger networks as well as improving model training
in a decentralized manner. Blockchain also might replace the aggregation server in cross-silo DL, which
means blockchain nodes can perform the global model aggregation task. Furthermore, blockchain with
SC features can be deployed as an incentive mechanism to motivate clients to collaborate on the global
model improvement using their local data.
3 Motivation
Most of the prior studies have been particularly focused on utilizing public blockchain platforms for
various use cases with transparency property. Some researches have either emphasized the merits of
blockchain smart contract [3, 29, 4], or utilized blockchain security as presented in [13, 25]. Blockchain
technology as an incentive mechanism with different platforms are also deployed in [35, 19, 42]. Nev-
ertheless, the additional security protocols that can be adjusted to the system requirements are rare to
be discussed, especially for sensitive data. The selected studies of previous researches in chronological
order can be seen in Table 1 that also summarizes the existing methods and highlights the importance of
our study in the matter of cross-silo distributed learning environment.
This research aims to bridge the gap by utilizing Monero (XMR) protocols in any decentralized trans-
actions activities. This research emphasizes the present state of the Ethereum blockchain smart contract
as a backbone technology in conducting decentralized transactions over peer-to-peer networks. Smart
contracts are self-executing contracts with irreversible data records that can be a plausible solution in
propagating incentives within distributed learning schemes. However, the two-phase commit designed
between users and model providers through the blockchain environment could violate users’ privacy
contrary to the principal objective of DL. Hence, by adopting inference attacks, the observer can infer
and link specific data features of the private dataset. This research investigates several pseudonymiza-
4
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Table 1: Summary of previous researches in chronological order.
Research Paper Year Objective Demerit Significance of the Study
Rahmadika and
Rhee [37]
2021 Untraceable transactions in the
cross-silo federated learning
Centralized aggregation server
(irreplacable)
Empirical benchmark
Rahmadika, Fir-
daus et al. [34]
2021 An intelligent cross-silo FL
based on distributed ledger
For general use cases only
(specific assumption)
Empirical and perfor-
mance benchmark
Shuaicheng et al.
[28]
2021 Privacy model for federated
learning running on BC
Impractical for several use
cases
Security and privacy
trade-offs
Ayaz et al. [3] 2021 BC and FL for message dis-
semination in VANETs
Linkability concerns in com-
munication exchanges
Unlinkable techniques
Mahmood and
Jusas [29]
2021 BC and FL adoption for classi-
fication problems
BC platform’s performances
are not defined
Performance benchmark
Qu and Wang et
al. [33]
2021 FL adoption as a PoW consen-
sus in blockchain
BC hardfork / radical changes
are required
Separation practices
benchmark studies
Feng et al. [13] 2021 Blockchain to provide security
in the MEC system
Transparency and untraceabil-
ity issues
Untraceable incentive
mechanism
Rahmadika and
Rhee [35]
2020 Reliable FL with blockchain-
based incentive
Impractical to be adopted in the
smart contract
Performance benchmark
and evaluation
Zhang et al. [46] 2020 BC with FL for device failure
detection in industrial IoT
Encryption techniques are not
elaborated
Privacy-preserving proto-
cols
Rehman et al.
[43]
2020 Blockchain-based reputation-
aware fine-grained FL (trust-
worthy FL in MEC system)
Impractical to be adopted in the
smart contract
Benchmark problems -
centralized aggregation
server/data centers
Khan et al. [19] 2020 Resource optimization and dis-
tributed reward
Linkable-reward mechanism Untraceability features
are required
Kumar et al. [25] 2020 Enhancing privacy in BC en-
abled FL
BC’s performances were not
elaborated
Performance benchmark
and evaluation
Bao and Su et al.
[4]
2019 A healthy marketplace with BC
and collaborative training
Undefined privacy-preserving
protocol and training auditing
Privacy preservation pro-
tocols
Kang et al. [17] 2019 Reliable FL reliable with mul-
tiweight suebjective logic
Privacy-awareness is beyond
the research
Empirical analysis
Toyoda et al. [42] 2019 An incentive-aware BC with
FL (incentive compatibility)
Key idea and design (theoreti-
cally)
Performance benchmark
& evaluation
tion techniques in obscuring decentralized transactions on the peer-to-peer network. This research also
outlines several points related to the concept, design and linkability issues of implemented schemes.
4 Privacy Awareness in Decentralized Approaches
First, this section delivers privacy awareness in decentralized approaches by stating the current state of
the blockchain-based decentralized learning environment. The existing contemporary works are also
presented. At the end of this section, the linkability concerns of decentralized learning with blockchain-
based incentive mechanisms are also discussed. We also highlight some research questions.
4.1 Blockchain-based Decentralized Learning
The conventional DL approach still faces several challenges that need to be addressed, especially in
privacy problems, such as membership inference attacks, data poisoning attacks, malicious clients, dis-
honest central aggregator servers, and the possibility of SPoF occurring. In the membership inference
5
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
attack, attackers might perform reverse engineering to gather the client’s private data by leveraging the
updated model training. In contrast, a poisoning attack affects the global model by sending the malicious
updated models during the collaborative training phase. Furthermore, the central aggregator responsi-
ble for managing whole system orchestration has trouble addressing crucial challenges associated with
the SPoF issue, which may cause the risk of client information being possibly exposed. As a result,
the clients could be reluctant to participate in improving the cross-silo DL system. On the other hand,
blockchain is an open database that guarantees data security by supporting trustworthy and anonymous
transactions without requiring any intermediaries. Moreover, those transactions are recorded on the dis-
tributed and immutable ledger [14]. Therefore, blockchain has also been exploited to tackle the several
flaws of conventional DL, as mentioned above. We investigate that the blockchain-based DL approach
has at least the following advantages:
(i) Blockchain can avoid SPoF and achieve decentralization by replacing the aggregation server (cen-
tralized approach) and allowing more than one blockchain node to execute the model aggregation
to be the global model in the cross-silo DL system.
(ii) The verification mechanism in the blockchain system can filter unreliable data from malicious
clients or other attacks (e.g., inference membership attack and data poisoning attack) before it is
stored and aggregated to be a global model. In this sense, only valid data will be aggregated to the
global model, while the unreliable data of local model updates will be detected in the verification
mechanism.
(iii) Blockchain can provide decentralized transactions, which means all transactions are marked with a
timestamp, and then a particular consensus mechanism validates and stores the verified transaction
on the distributed database network. Hence, the system allows all involved participants in the
blockchain network to obtain an updated ledger automatically.
(iv) Blockchain with SC can be deployed to address the lack of incentives as one of the flaws in con-
ventional cross-silo DL. Here, blockchain can be utilized to distribute incentives (i.e., rewards) to
clients. Thus, the incentive mechanism can encourage the clients to honestly contribute to training
their data using their computational resources to improve the global model.
The blockchain-based DL has been mentioned in various current existing works. In [27], the BLADE-
FL framework is proposed, which aims to deploy the fully decentralized model aggregation. Moreover,
this framework provides a reliable learning environment by encouraging a self-motivated approach for
clients. The smart contract is designed to integrate clients’ mining and training tasks to calculate and
update a global model. BlockFL architecture is proposed by Kim et al. [21] that focuses on providing
a decentralized manner by removing the aggregation server. BlockFL, as the combination of blockchain
and FL, is used to verify and exchange the local model updates generated by clients without coordination
from a centralized server. Figure 2 illustrates the framework of conventional DL and blockchain-based
DL. In general, as shown in Figure 2(b), blockchain-based DL allows the client to perform local model
training and verify the updated models to generate the global model in a decentralized manner. Weng
et al. [44] proposed DeepChain as a fair, secure, and distributed protocol by providing an incentive
mechanism based on blockchain to motivate clients to behave correctly in the system. DeepChain pro-
tocol requires every user to state their asset to access the system and perform their task to train the DL
model collaboratively. In this regard, users UE xnsend the asset transaction T x U Exn(Asset )using their
pseudonymous address public key PK UE xnto prove their asset U ExnAsset ownership (see formula
6
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Figure 2: (a) Conventional decentralized learning; (b) Blockchain-based decentralized learning.
[44]).
T x U E xn(Asset)=PK UExn→PKUExnAsset =PubKeyHash(UE xnAsset),
αjUExn=Hash(j).PubKeyHash(UE xnAsset)H ash(UE xnAsset),”Asset desc”,
where PK U Exn∈PubKey1Sec UE x ,PubKey2Sec U Ex, ...,PubKeynSec U E x
(1)
Formula 1 describes the asset statement transaction of UExn. It consists of the client’s asset pseudonym
public key PKUE xnAsset , the proof of UEx’s asset ownership αjUExnand the U E xn’s asset descriptions
”Asset desc” (e.g., data topic, data format, and data size). Here, PKUE xnAsset and αjUE xnare essential
components that can be utilized to prove the client’s asset ownership and ensure that UExnAsset can
not be revealed. These asset statement components are composed of the collision-resistant hash func-
tion for mapping UE x’s assets Hash(UExnAsset)and the unique-generated public keys PubKey1Sec UE x
with corresponding private keys to maintain pseudonymity. In order to form a fair and secure collabo-
rative transaction, DeepChain proposed a collaborative information commitment including the number
of clients U Exn, index of the current iteration t, the parameter of threshold using Threshold Paillier
algorithm T hresMDL, client’s commitment Commit.Sec , collaborative global model φglb(x), the initial
weights W0,jand the amount of client’s deposit UExcoin. In short, all those information are recorded in
a collaborative transaction and associated with being a collective address PKcollab before uploading it to
DeepChain [44].
T x(collab)=PKcollab →UExn,t,T hres(MDL),Commit.Sec ,φglb(x),W0,j,UE xcoin (2)
The fundamental DL model with blockchain-based revenue is shown in Algorithm 1. The high-
level model is divided into three procedures: server update mechanism, user update, and blockchain
revenue. In the server update mechanism, the model provider estimates the number of potential users
Poss.UExnto be included to build the model. Several requirements need to be met, such as device type,
7
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Algorithm 1 Fundamental DL technique with BC-based revenue. BSz is the global batch size; UExnare
users indexed by i; where Min BSz is local minibatch size, Lrate is learning rate, inspired by [23].
1: procedure SERVE RUPDATE MECHANISM:
2: Aggsvr roughly mapping Poss.UExn*aggregation server initialized the potential users
3: for each round t= 1,2, ..., n do
4: Poss.UExnglb ←maximum(BSz ·UExnPoss.(i,j))
5: Poss.UExn←(subset Poss.UE xnglb users) *the users are a subset of total global U Exnavailable
6: for each user i∈Poss.UExnin parallel do
7: Poss.UExni
t←UserUpdate(i,Poss.UExnt)*∀updated model is based on Poss.UExnglb
8: end for
9: Poss.UExnu←1
nt ∑i∈Poss.U ExnnPoss.UExni
u
10: Aggsvr ←Poss.UExnt+1=Poss.U Exn +LratePoss.UE xnu*gathering updated gradient
11: end for
12: end procedure
13: procedure USERUPDATE:(i,Poss.UExn)*the updated model is broadcasted to the users
14: //Executes on user i
15: Min BSz ←(split data iinto batches of size Min BSz)
16: for every local epoch nt from 1 to Edo
17: for batch Min BSz1∈Min BSz total do
18: Poss.UExni(MDL)←Poss.UExni
t−Poss.UExnt;*∀UExnreceive the updated model
19: end for
20: end for
21: return Poss.UExn list to aggregation server Aggsvr
22: end procedure
23: procedure ETHE R REVE NUE(Rv UExn,Rvmg)*incentivized using Ethereum platform
24: Aggsvr collects the list of sender (users) UExn1,U E xn2, ...,U E xn
25: Active miners mg1,mg2,..., mgn
26: for mg1,mg1,...,mgn∈Minertot ;Aggsvr do
27: Aggsvr ←ConfirmTransaction H(Poss.UE xn1
t,Poss.UExn2
t,..., Poss.UExnnt
t)
28: *Aggsvr has the list of users
29: *Other miners validate the result till it gets confirmed
30: Rv UExnare given to UExn1,UExn2, ...,U E xn*the rewards are distributed to the users
31: Rvmg are distruted to mg1,mg1,..., mgn∈Minertot *mining reward for the miners
32: end for
33: end procedure
the network latency, the minimum number of the dataset, and to name a few. The users send the gradient
model back and forth to the model provider after finishing the private training using their dataset. The
users are rewarded through the Ethereum blockchain whenever the UE x’s transactions are satisfied the
requirements. The revenue Rv U E xnbelongs to the data owner, and the revenue Rvmg is mining revenue
for the miners (automatically distributed). We detail this point in Section 5.
4.2 Linkability Concerns at a First Glance
The merits of Ethereum smart contracts can be utilized as a rewarding platform in the DL scheme. It
is irreversible and tamper-proof that can solve the dispute between parties. This section illustrates an
incentive mechanism in distributed learning by leveraging the Ethereum smart contract to tackle the
intermediaries issues in the centralized incentive schemes. Nevertheless, concerns arise if this scheme is
implemented for sensitive data such as health-related data since the gradient values [24] from users are
exposed publicly through the smart contract. The observer can adopt the active and passive inference
attack with certain assumptions to impose training data through gradients value that breaks privacy. We
8
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Figure 3: (a) Accuracy performance of DL learning; (b) The average amount of gas used by devices and
smart contract manager; (c) Inferring the users’ information.
suggest the reader refer to our prior work in [36] to comprehend the DL model setting details. Other
similar attacks and concerns are also described in [40, 10] .
The users are data owners who possess a large amount of private training data, while the model
provider preserves the deep learning model that can be downloaded publicly. The gradient values are
gathered by the provider gradually, which later to be used to compute the aggregation value. The provider
provides Agsvr revenue through smart contracts SCxfor those proven to contribute to improving the
models (using users’ resources). The communication between providers and users is carried out via
two-phase commit transaction. The performance results can be seen in Figure 3 (a) and (b). The overall
results positively recommend that the schemes can be applied to real-world implementation (for non-
sensitive data). However, when the users Dxndeploy the transaction that consists of a cipher to encrypt
the information to the provider, the observer can impose the dataset knowledge by adopting active and
passive inference attacks [30] as shown in Figure 3 (c). Yet, the performance of the adversary decreases
with an increasing number of users. Blockchain can be utilized even further as a decentralized gradient
counter. In this sense, the role of the centralized aggregation server can be replaced with distributed
nodes scattered in the network.
5 Secure Decentralized Transactions
This section presents the techniques in securing decentralized learning transactions where the entities
can securely conduct several activities. The entities’ information remains secret. We also investigate the
pseudonymous rewarding mechanism by utilizing the XMR protocols. In the final section, comparative
analysis are discussed.
5.1 Secure Decentralized Learning Transactions
Privacy preservation and anonymity are the most paramount aspects of decentralized transaction activ-
ities. Peer-to-peer transactions seek to be concealed from the observer’s view, and a distinct difference
resembled the centralized approach. In particular, CryptoNote protocol stated two properties that a fully
anonymous cryptocurrency standard must meet to comply with the requirements presented by Okamoto
and Ohta [31] as follows: (i) Untraceable transactions. For the respectively incoming transactions, the
probability of all potential senders is equiprobable. (ii) Unlinkable transactions. It is extremely hard to
prove whether the same entity sent the two or more outgoing transactions.
9
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Perversely, most decentralized transactions activities from many different platforms do not satisfy
both requirements, especially the untraceability property. Every transaction can be unambiguously traced
to a unique origin and final recipient since the transactions are conducted on the public network. Even
when entities indirectly exchange the data or funds, adequately engineered path-finding algorithms will
expose the source and final recipient. In regards to the cryptocurrency as a reward in DL activities, Bit-
coin does not meet the unlikability property. A sophisticated blockchain analysis can unveil a relationship
between the users of the Bitcoin network and their associated transactions. Therefore, several techniques
have emerged to answer these challenges, such as Monero (XMR), which applies the CryptoNote proto-
col as the backbone of the cryptocurrency. In this research, we utilize some properties of the CryptoNote
protocol to support the pseudonymous of decentralized learning activities.
The first feature begins with modifying the ring signature algorithm, where BPx be the ed25519
basepoint as part of Edwards-curve digital signature algorithms using secure hash algorithms 512 (SHA-
512) and a Curve25519 (q=2255 −19)with a twisted Edwards curve shown in formula (3). For detailed
information, we recommend readers refer to fast explicit formulas techniques for group operations on an
Edwards curve, elaborated in [6].
−CorX2+CorY 2=1−121665
121666 CorX 2·CorY 2,
W here l =2252 +2774231777372353 ... and c =3
CorX =ValU
ValV √−4886664,with CorY =ValU −1
ValU +1
(equivalent t o the Montgomery curve)
(3)
We note that every hash value produces a point in accumulating the base point BPx (Hash =ψBPx for
any undefined ψ). Contrary to what occurs in secp256k1 (the curve is leveraged in the Bitcoin cryptocur-
rency). Suppose Commit.(a,CorX) = CorX ·BPx +a·H ash, the commitment to the value awith mask
CorX . We realize that as long as logBPx Hash is defined, while a6=0, then logBPx Commit.(a,CorX )is
remain unspecified. Contrary, with the value of a=0, then logBPx Commit.(a,CorX ) = CorX . In this
sense, it is possible to sign with the secret of sender’s private key. Eventually, the networks can check
whether the input commitments and output commitments are ∑In puts =∑Out puts. Yet, these proper-
ties are not sufficient in XMR because the given transactions TX s consists of multiple potential inputs
Poss.i,i=1,2,3, ..., n, where only one of which corresponds to the sender. This concern is not expected
because it eliminates the anonymity produced by the ring signatures protocol. Hence, the commitments
are constructed in (4) as follows:
Commit.in puts =CorXcommit.·BPx +a·H ash
Commit.(out put −1)=CorY1·BPx +b1·Hash
Commit.(out put −2)=CorY2·BPx +b2·Hash
Commit.(out put −n)=CorYn·BPx +bn·Hash
(4)
The constructed ring signatures protocol consists of all Commit.i,i=1,sec, ..., nwith sec is the
secret key index of the commitment of the sender, defining the corresponding public key. In this sense,
the commitments and public keys are paired (Commit.i,Poss.i); while the subtracting ∑Commit.out is
generated in advance. In this case, the commitments are designed in formula (5). This formula is a ring
signature that can be signed since the senders have knowledge of the private keys. Precisely, due to the
knowledge of private key for Poss.iand the private key for Poss.i+Commit.i,in −∑jCommit.j,out , the
sender can conduct a signature for any desired transactions. The signer can use the formula (5) to sign a
10
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
transaction, as we defined in our previous work in [37].
Poss.(1)+Commit.(1,in)−∑
j
Commit.(j,out ), ..., Poss.(sec)+Commit.(sec,in)
−∑
j
Commit.(j,out ), ..., Poss.(n) + Commit.(n,in)−∑
j
Commit.(j,out ).
(5)
DL T xφ1req.
•RNGsgn(i,j)∈RN Gtot ≡v”AND”
”T xφnreq.”=True →then”Approve”; Otherwise :
•RNGsgn(i,j)/∈RNGtotv ”OR”
”T xφnreq”=False →t hen”Decline”
(6)
For ease of understanding, we construct a scenario where the users desire to use a DL model φglb(x)
uploaded onto cloud services. The users are required to deploy a transaction request that states the desired
global model along with relevant information (the model owner governs the necessary inputs). The users
select the number of signatures to be applied to sign a request transaction DL T xφ1req.; where this is
can be understood as a pseudonymous request transaction with an anonymity feature. The model owner
can frequently change the format of transactions, yet the protocols behind the transactions remain the
same. When the model owner receives the DL T xφ1req.sent by users via a secure channel, then the
owner checks the condition of the transaction to confirm the completeness of requirements as defined
in (6). In the first place, the model owner checks the correctness of the ring signatures that he received
beforehand (it belongs to the group or not). The owner repudiates the transactions if one of the conditions
is not satisfied. The models are sent to the users if only conditions in (6) are met.
5.2 Pseudonymous Rewarding Mechanism and Comparative Analysis
In practice, most incentive schemes are centralized where the third party has a vital role in conducting
transactions. The centralized nature is inherent to the single point of failure (SPoF) [39] and bottleneck
issues that jeopardize the system’s root. On the other hand, with its merits, blockchain can be a plausible
solution to tackle the problems since it runs on top of the peer-to-peer network where no single interme-
diary has complete control over transactions. Accordingly, blockchain-based reward mechanisms have
been heavily adopted in many various use cases. For instance, Lin et al. [26] utilized blockchain to
deliver rewards in the energy-knowledge trading environment. In comparison, Chakrabarti et al. [9] pro-
posed a blockchain-based reward scheme for opportunistic disaster communication over a delay tolerant
network. Nevertheless, straightforwardly applying blockchain-based incentives for sensitive informa-
tion is not desirable. This section relies on several XMR protocols’ features to support pseudonymous
rewarding in decentralized learning.
The users are rewarded since they build the DL model collaboratively by broadcasting the updated
gradient values to the model provider. To be incentivized, the sender generates a Diffie-Hellman key
exchange protocol to obtain a shared secret key from his data and half of the recipient’s address. The
users UEx(i,j)are also needed to tender a new transaction via a private Ethereum smart contract. This
transaction is denoted as DL T xφnRwd.that proves the user’s contribution in building a DL model
using their corresponding valuable datasets. The DL T xφnRwd.transaction has a distinguish feature
with DL T xφ1req.transaction, where within DL T xφnRwd.there are a pair of public keys (PubA and
PubB). The first public key PubA is performed by the UE xnusing U Exn’s private key SecA blended with
a base point BPxA;PubA →SecA ·BPxA. Similarly, the other public key PubB is generated from a private
key SecB with their respective generator BPxB;PubB →SecB ·BPxB. The produced public key must be
11
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Figure 4: Structure of standard transactions.
unique since the private keys are different from others since the generators are also unique SecA 6=SecB
”AND” BPxA6=BPxB.
DL T xφnRwd.=
φglb(in f o)||φu pδ1||δ1knowledge
UExn0s PubKey →PubA,PubB (Generated)
{RNGsgn(i,j)∈RN Gtot ≥1||SecA} → Signed
(7)
Dest.ψ=BPxn+U E XnPubB ·Hash(random UEXnSecA)(8)
Spendψ=S ecB +Hash(SecA ·rand om ·BPXn)(9)
Figure 5: The illustration of distribution points of time spent by UExnto generate a shared public key.
This process is carried out for 50 repetitions for each number of RNGsgn .
Formula (7) represents the UExn’s reward transaction. It consists of the global model of DL’s infor-
mation φglb(in f o), the updated gradient values φu pδ1, and the knowledge of dataset δ1knowledge. The
parameters of the transaction can be adjusted accordingly. The user UExnalso attached a pair of public
keys (PubA and PubB)within transactions which are derived from SecA 6=SecB ”AND” BPxA6=BPxB.
12
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Suppose random =Owner0s random data →R=random·BPxnis part of Diffie Hellman exchange prin-
ciple. If DL T xφnRwd .condition is satisfied, the DL model provider can unpack the public key attached
to the transaction. The provider then generates another random base point rb p ∈[1,l−1]and manages
the one-time destination key Dest.ψas depicted in 8 to be sent back to the UE xnvia Ethereum smart
contract. The U E xnas a recipient checks every passing blockchain transaction using his private key SecA
and SecB.UExnwill be able to recover the respective one-time private key to spend the Ether since only
UExnknows about SecA and SecB as shown in formula (9). In short, UExnis the only legitimate user.
Table 2: Ethereum daily gas used information (historical total daily).
No. Date (UTC) POSIX time Total Value
1 Jan. 1st, 2021 1609459200 80034402241
2 Feb. 1st, 2021 1612137600 79242528840
3 Mar. 1st, 2021 1614556800 79006491494
4 Apr. 1st, 2021 1617235200 79787949899
5 May. 1st, 2021 1619827200 94431457868
... ... ... ...
9 Sep. 1st, 2021 1630454400 99646996226
10 Oct. 1st, 2021 1633046400 99940537596
Figure 6: The sequence number of 50 shared key generations for each different RNGsgn.
In terms of time spent in generating a shared key for entities is depicted in Figure 5 (XMR-based
protocols). This illustration figure is representative of distribution points of time spent by U E xnin
performing a shared public key to be used to conduct a transaction. For another perspective view, the
transactions in sequential order can be seen in Figure 6. This process is carried out for 50 repetitions
with a varying number of UExn. Suppose user 1 UEx1gathers another public key to create a group
ring of signature. First, UE x1selects one public key of UE x2. He also chooses the other two public
keys (UEx1,U E x2,U Ex3). Finally, for the last transaction, UEx1uses four public keys including himself
13
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Table 3: Performance benchmark with several existing approaches.
Study Year BC Platform Concerns Addressed Unlink. TXs
Our approach 2021 Ethereum Unlinkability TXs and privacy preserv. Yes (Prot.)
Feng et al. [13] 2021 Hyperledger Security and trust issues over MENs Yes (Part.)
Ma et al. [28] 2021 SCx-based Transparency and contribution evaluation N/D
Ayaz et al. [3] 2021 SCx-based Message dissemination in VANETs N/D
Fan et al. [12] 2020 FISCO-BCOS Auditable rational reverse auction N/D
Sandi et al. [35] 2020 Ethereum Commensurate decentralized incentive N/D
Khan et al. [19] 2020 N/D Incentive-based entities interaction N/D
Weng et al. [44] 2019 Corda Decentralized incentive mechanism Yes (Part.)
Prototype (Prot.); Partial (Part.); Not defined (N/D);
(UE x1,U Ex2,U Ex3,and U Ex4).
The average generating time for the first creation (UEx1andU E x1) was 66.415ms, with the fastest
and longest time were 55.876ms and 76.263ms, respectively. The same procedures are applied to the sec-
ond creation, where U E x1selects U Ex1,U Ex2,U Ex3keys. The average generating time was recorded at
74.103ms, with the fastest time was 71.133ms, and the longest time was recorded at 78.274ms. Eventu-
ally, the last recorded generation time of the shared key is carried out by selecting four keys of the users.
The average time was 84.768ms, with the fastest and the longest time were 82.776ms and 86.055ms. The
simulation results show that the more keys involved, the more time it takes to create a shared key. This
concern becomes essential to be taken into account since it directly affects the gas usage in the Ethereum
smart contract transactions. The amount of gas usage continues to increase over time, affecting the gas
limit and gas price. In the end, the uncertainty of the Ethereum gas affects the cost fee per transaction.
Therefore, the effective design of smart contracts is essential. The increasing number of Ethereum gas
usage can be seen in Table 2 that describes the Unix time-stamps and the values. The data was recorded
from January 1st, 2021, up to October 1st, 2021 [38]. We also recommend readers refer to our previous
works in [34] and [37].
Finally, we emphasize the unlinkability feature in decentralized learning transactions with several
existing approaches, as highlighted in Table 3. We limit the transactions into two categories, namely,
a request transaction and distributed reward transaction. Our proposed scheme covers both transactions
where also can be adjusted by following the design requirements. However, we only embed the protocols
into smart contract transactions in a prototype form due to the hard fork concerns that require radical
changes to the entire Ethereum networks. The previous work in [13] and [44] also provided some parts of
privacy techniques in decentralized transactions activity. However, the precise methods of unlinkability
features were not described in the paper. In contrast, other works listed in Table 3 directly adopted the
blockchain and decentralized learning without addressing the linkability issues. This research fills the
current literature gap concerning a recent systematic mapping study to preserve privacy in decentralized
transactions.
6 Conclusion
We have presented and investigated the pseudonymization techniques in decentralized transactions. Cross-
silo distributed learning and blockchain smart contracts emerged as backbone technologies to tackle sev-
eral issues in the centralized system. This investigation becomes essential to be discussed since many
existing schemes only provide privacy protocols, yet the linkability concerns are beyond the topics.
14
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Thus, we utilized the XMR protocols to be adopted into the DL system with a blockchain-based incen-
tive mechanism that can cover many centralized issues. Simulation results and performance benchmarks
indicate that the XMR protocols can be a plausible solution to address the linkability concerns inher-
ent to the public blockchain in general. In order to be fully implemented into the Ethereum network,
a radical change (hard-fork) is required that makes predecessor invalid blocks and transactions become
valid and vice versa. Therefore, we limit this research by implanting the designed protocols into smart
contract transactions, where the protocols can be adjusted proportionately. Apart from the benefits of the
proposed scheme, several points need to be explored further, such as the aggregation server involvement
(irreplaceable), the amount of gas usage in the Ethereum, the devices availability, and to name a few.
These points are part of our research interest for the long run.
Acknowledgments
This research was supported in part by the MSIT(Ministry of Science and ICT), Korea, under the
ITRC(Information Technology Research Center) support program(IITP-2021-2020-0-01797) supervised
by the IITP(Institute for Information & Communications Technology Planning & Evaluation), and in part
by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded
by the Ministry of Education (2021R1I1A304659011).
References
[1] Blockchain table. https://docs.oracle.com/en/database/oracle/oracle-database/ [Online; ac-
cessed on September 20, 2021].
[2] T. T. Anh, N. C. Luong, D. Niyato, D. I. Kim, and L.-C. Wang. Efficient training management for mobile
crowd-machine learning: A deep reinforcement learning approach. IEEE Wireless Communications Letters,
8(5):1345–1348, October 2019.
[3] F. Ayaz, Z. Sheng, D. Tian, and Y. L. Guan. A blockchain based federated learning for message dissemination
in vehicular networks. arXiv preprint arXiv:2109.06667, September 2021.
[4] X. Bao, C. Su, Y. Xiong, W. Huang, and Y. Hu. Flchain: A blockchain for auditable federated learning with
trust and incentive. In Proc. of the 5th International Conference on Big Data Computing and Communications
(BIGCOM’19), Shandong Province, China, pages 151–159. IEEE, August 2019.
[5] D. Berbecaru, A. Lioy, and C. Cameroni. Supporting authorize-then-authenticate for wi-fi access based
on an electronic identity infrastructure. Journal of Wireless Mobile Networks, Ubiquitous Computing, and
Dependable Applications, 11(2):34–54, June 2020.
[6] D. J. Bernstein and T. Lange. Faster addition and doubling on elliptic curves. In Proc. of the 2007 Interna-
tional Conference on the Theory and Application of Cryptology and Information Security (ASIACRYPT’07),
Kuching, Malaysia, volume 4833 of Lecture Notes in Computer Science, pages 29–50. Springer, Berlin,
Heidelberg, December 2007.
[7] A. N. Bhagoji, S. Chakraborty, P. Mittal, and S. Calo. Analyzing federated learning through an adversarial
lens. In Proc. of the 36th International Conference on Machine Learning (ICML’19), California, USA,
volume 97, pages 634–643. PMLR, June 2019.
[8] D. Caputo, L. Verderame, A. Ranieri, A. Merlo, and L. Caviglione. Fine-hearing google home: why silence
will not protect your privacy. Journal of Wireless Mobile Networks, Ubiquitous Computing, and Dependable
Applications, 11(1):35–53, March 2020.
[9] C. Chakrabarti and S. Basu. A blockchain based incentive scheme for post disaster opportunistic commu-
nication over dtn. In Proc. of the 20th International Conference on Distributed Computing and Networking
(ICDCN’19), Bangalore, India, pages 385–388. ACM, January 2019.
[10] R. Chen, J. Guo, D.-C. Wang, J. J. Tsai, H. Al-Hamadi, and I. You. Trust-based service management for
mobile cloud iot systems. IEEE transactions on network and service management, 16(1):246–263, December
2018.
15
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
[11] A. De Villiers and P. Cuffe. A three-tier framework for understanding disruption trajectories for blockchain
in the electricity industry. IEEE Access, 8:65670–65682, March 2020.
[12] S. Fan, H. Zhang, Y. Zeng, and W. Cai. Hybrid blockchain-based resource trading system for federated
learning in edge computing. IEEE Internet of Things Journal, 8(4):2252–2264, October 2020.
[13] L. Feng, Z. Yang, S. Guo, X. Qiu, W. Li, and P. Yu. Two-layered blockchain architecture for federated
learning over mobile edge network. IEEE Network, Early Access:1–14, January 2021.
[14] M. Firdaus, S. Rahmadika, and K.-H. Rhee. Decentralized trusted data sharing management on internet of
vehicle edge computing (iovec) networks using consortium blockchain. Sensors, 21(7):2410, March 2021.
[15] A. Hard, K. Rao, R. Mathews, S. Ramaswamy, F. Beaufays, S. Augenstein, H. Eichner, C. Kiddon, and
D. Ramage. Federated learning for mobile keyboard prediction. arXiv preprint arXiv:1811.03604, November
2018.
[16] R. Kanagavelu, Z. Li, J. Samsudin, Y. Yang, F. Yang, R. S. M. Goh, M. Cheah, P. Wiwatphonthana,
K. Akkarajitsakul, and S. Wang. Two-phase multi-party computation enabled privacy-preserving federated
learning. In Proc. of the 20th IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing
(CCGRID’20), Melbourne, Victoria, Australia, pages 410–419. IEEE, May 2020.
[17] J. Kang, Z. Xiong, D. Niyato, S. Xie, and J. Zhang. Incentive mechanism for reliable federated learning: A
joint optimization approach to combining reputation and contract theory. IEEE Internet of Things Journal,
6(6):10700–10714, September 2019.
[18] J. Kang, Z. Xiong, D. Niyato, Y. Zou, Y. Zhang, and M. Guizani. Reliable federated learning for mobile
networks. IEEE Wireless Communications, 27(2):72–80, February 2020.
[19] L. U. Khan, S. R. Pandey, N. H. Tran, W. Saad, Z. Han, M. N. Nguyen, and C. S. Hong. Federated learning
for edge networks: Resource optimization and incentive mechanism. IEEE Communications Magazine,
58(10):88–93, November 2020.
[20] L. Kiffer, D. Levin, and A. Mislove. Analyzing ethereum’s contract topology. In Proc. of the 2018 Inter-
net Measurement Conference 2018 (IMC’18), Boston, Massachusetts, USA, pages 494–499. ACM, October
2018.
[21] H. Kim, J. Park, M. Bennis, and S.-L. Kim. Blockchained on-device federated learning. IEEE Communica-
tions Letters, 24(6):1279–1283, June 2019.
[22] A. Koloskova, S. U. Stich, and M. Jaggi. Decentralized stochastic optimization and gossip algorithms with
compressed communication. arXiv preprint arXiv:1902.00340, February 2019.
[23] J. Koneˇ
cn`
y, H. B. McMahan, F. X. Yu, P. Richt´
arik, A. T. Suresh, and D. Bacon. Federated learning:
Strategies for improving communication efficiency. https://arxiv.org/abs/1610.05492, October 2016.
[24] P. Kumar, S. Garg, A. Singh, S. Batra, N. Kumar, and I. You. Mvo-based 2-d path planning scheme for
providing quality of service in uav environment. IEEE Internet of Things Journal, 5(3):1698–1707, January
2018.
[25] S. Kumar, S. Dutta, S. Chatturvedi, and M. Bhatia. Strategies for enhancing training and privacy in blockchain
enabled federated learning. In Proc. of the 6th IEEE International Conference on Multimedia Big Data
(BigMM’20), New Delhi, India, pages 333–340. IEEE, September 2020.
[26] X. Lin, J. Li, J. Wu, H. Liang, and W. Yang. Making knowledge tradable in edge-ai enabled iot: A con-
sortium blockchain-based efficient and incentive approach. IEEE Transactions on Industrial Informatics,
15(12):6367–6378, May 2019.
[27] C. Ma, J. Li, M. Ding, L. Shi, T. Wang, Z. Han, and H. V. Poor. When federated learning meets blockchain:
A new distributed learning paradigm. arXiv preprint arXiv:2009.09338, September 2020.
[28] S. Ma, Y. Cao, and L. Xiong. Transparent contribution evaluation for secure federated learning on blockchain.
In Proc. of the 37th IEEE International Conference on Data Engineering Workshops (ICDEW’21), Chania,
Greece, pages 88–91. IEEE, April 2021.
[29] Z. Mahmood and V. Jusas. Implementation framework for a blockchain-based federated learning model for
classification problems. Symmetry, 13(7):1116, June 2021.
[30] L. Melis, C. Song, E. De Cristofaro, and V. Shmatikov. Exploiting unintended feature leakage in collaborative
learning. In Proc. of the 2019 IEEE Symposium on Security and Privacy (SP’19), San Francisco, California,
USA, pages 691–706. IEEE, May 2019.
16
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
[31] T. Okamoto and K. Ohta. Universal electronic cash. In Proc. of the 1991 Annual international cryptol-
ogy conference (CRYPTO’91), Santa Barbara, California, USA, volume 576 of Lecture Notes in Computer
Science, pages 324–337. Springer, Berlin, Heidelberg, May 1991.
[32] M. Park, S. Kim, and J. Kim. Research on note-taking apps with security features. Journal of Wireless Mobile
Networks, Ubiquitous Computing, and Dependable Applications, 11(4):63–76, December 2020.
[33] X. Qu, S. Wang, Q. Hu, and X. Cheng. Proof of federated learning: A novel energy-recycling consensus
algorithm. IEEE Transactions on Parallel and Distributed Systems, 32(8):2074–2085, August 2021.
[34] S. Rahmadika, M. Firdaus, S. Jang, and K.-H. Rhee. Blockchain-enabled 5g edge networks and beyond: An
intelligent cross-silo federated learning approach. Security and Communication Networks, 2021:1–14, March
2021.
[35] S. Rahmadika and K.-H. Rhee. Reliable collaborative learning with commensurate incentive schemes. In
Proc. of the 2020 IEEE International Conference on Blockchain (Blockchain’20), Rhodes, Greece, pages
496–502. IEEE, November 2020.
[36] S. Rahmadika and K.-H. Rhee. Enhancing data privacy through a decentralised predictive model with
blockchain-based revenue. International Journal of Ad Hoc and Ubiquitous Computing, 37(1):1–15, May
2021.
[37] S. Rahmadika and K.-H. Rhee. Unlinkable collaborative learning transactions: Privacy-awareness in decen-
tralized approaches. IEEE Access, 9:65293–65307, April 2021.
[38] E. Scan. Ethereum daily gas used chart. https://etherscan.io/chart/gasused [Online; accessed on
October 15, 2021], October 2021.
[39] V. Sharma, I. You, and G. Kul. Socializing drones for inter-service operability in ultra-dense wireless net-
works using blockchain. In Proc. of the 2017 international workshop on managing insider security threats
(MIST’17), Dallas, Texas, USA, pages 81–84. ACM, October 2017.
[40] V. Sharma, I. You, K. Yim, R. Chen, and J.-H. Cho. Briot: Behavior rule specification-based misbehavior
detection for iot-embedded cyber-physical systems. IEEE Access, 7:118556–118580, May 2019.
[41] R. Shokri, M. Stronati, C. Song, and V. Shmatikov. Membership inference attacks against machine learning
models. In Proc. of the 2017 IEEE Symposium on Security and Privacy (SP’17), San Jose, California, USA,
pages 3–18. IEEE, May 2017.
[42] K. Toyoda and A. N. Zhang. Mechanism design for an incentive-aware blockchain-enabled federated learning
platform. In Proc. of the 2019 IEEE International Conference on Big Data (Big Data’19), Los Angeles,
California, USA, pages 395–403. IEEE, December 2019.
[43] M. H. ur Rehman, K. Salah, E. Damiani, and D. Svetinovic. Towards blockchain-based reputation-aware
federated learning. In Proc. of the 2020 IEEE Conference on Computer Communications Workshops (INFO-
COM WKSHPS’20), Virtual Conference, pages 183–188. IEEE, July 2020.
[44] J. Weng, J. Weng, J. Zhang, M. Li, Y. Zhang, and W. Luo. Deepchain: Auditable and privacy-preserving
deep learning with blockchain-based incentive. IEEE Transactions on Dependable and Secure Computing,
14(8):1–18, November 2019.
[45] T. Yang, G. Andrew, H. Eichner, H. Sun, W. Li, N. Kong, D. Ramage, and F. Beaufays. Applied federated
learning: Improving google keyboard query suggestions. arXiv preprint arXiv:1812.02903, December 2018.
[46] W. Zhang, Q. Lu, Q. Yu, Z. Li, Y. Liu, S. K. Lo, S. Chen, X. Xu, and L. Zhu. Blockchain-based federated
learning for device failure detection in industrial iot. IEEE Internet of Things Journal, 8(7):5926–5937,
October 2020.
——————————————————————————
17
Pseudonymization Techniques in Decentralized Transactions Sandi and Muhammad et al.
Author Biography
Sandi Rahmadika received the Ph.D. degree from Pukyong National University,
South Korea. He is currently a Postdoctoral Researcher with the Department of Dig-
ital Contents Engineering, Wonkwang University, South Korea. He has authored/co-
authored more than 25 papers in academic conferences and journals. He was the
Track Chair of the International Symposium and Mobile Internet Security (MobiSec)
2021. His research interests include applied cryptography, privacy preservation in the
decentralized system, edge computing, and AI with blockchain integration.
Muhammad Firdaus received his Master of Engineering degree in Telematics and
Telecommunication Networks from Institut Teknologi Bandung (ITB), Indonesia. He
is currently a Ph.D. student and a member of the Laboratory of Information Secu-
rity and Internet Applications (LISIA), Pukyong National University. His research
interests include applied cryptography, blockchain with AI integration, and commu-
nication security.
Yong-Hwan Lee received the MS degree in computer science and PhD in electronics
and computer engineering from Dankook University, Korea, in 1995 and 2007, re-
spectively. He is an active member of International Standard committees of ISO/IEC
JTC1 SC29 responsible for Image Retrieval and Coding issues. Currently, he is a
Professor at the Department of Digital Contents, Wonkwang University, Korea. His
research areas include Image Retrieval, Image Coding, Computer Vision and Pattern
Recognition, Augmented Reality, Mobile Programming and Multimedia Communi-
cation.
Kyung-Hyune Rhee received his M.S. and Ph.D. degrees from the Korea Advanced
Institute of Science and Technology (KAIST), Republic of Korea in 1985 and 1992,
respectively. He worked as a senior researcher at the Electronic and Telecommuni-
cations Research Institute (ETRI), Republic of Korea, from 1985 to 1993. He also
worked as a visiting scholar at the University of Adelaide, University of Tokyo, and
the University of California, Irvine. He has served as a Chairman of the Division of
Information and Communication Technology, Colombo Plan Staff College for Tech-
nician Education in Manila, the Philippines. He is currently a professor in the Department of IT Con-
vergence and Application Engineering, Pukyong National University, Republic of Korea. His research
interests center on security and the evaluation of blockchain technology, key management and its appli-
cations, and AI-enabled security evaluation of cryptographic algorithms.
18
