Federated Learning (FL) is a distributive machine learning approach for privacy-preserving sensitive data. FL uses a server to coordinate, initiate distributive learning and generate the aggregated model from the outcome of distributed nodes. Each collaborative distributive node generates a local model or portion of the model by tanning from its locally available data. The locally trained model weights are communicated to the server for generating the aggregated final model. The server shares the aggregated final model with the collaborating nodes to use. During those weight communication and model communication, FL algorithms sufer from the attack of malicious data. Generally, federated learning can use encryption or other defined techniques to ensure data privacy or security. But encryption makes more computational complexity and implementation complexity in the large distributed system. A blockchain is a distributed public digital ledger that generates blocks for every transition and stores them in all the computers in the network. The block of record cannot be altered without altering the content in all computers in the network's consensus. This distributive nature and immutability of transactions of blockchain enable the Integration of Blockchain with FL for securing the trained models' integrity. This research presents a comprehensive survey of blockchain-integrated Federated Learning.
Federated Learning is alternatively referred to as
collaborative learning is a kind of machine learning that
involves the training of an algorithm over a network of
dispersed edge devices or servers that store local data
slices without swapping them. Federated learning allows
several parties to construct a strong machine learning
model with no information sharing, there by
addressing crucial concerns such as data security, data privacy,
access rights and access to diverse data. Its use cases
are spread in a wide variety of business, including
pharmaceutics, defense, IoT, and telecommunications. The
attractive characteristics of Federated learning [
There are a few research challenges [
Federated learning distributive learning uses sensitive data such as patient health data, safety-related industrial data, and personal banking information collected via the Internet of Things (IoT) applications which need to be secured and the privacy of data to be maintained while training and testing of the machine learning models.
Figure 1 shows the typical federated machine learning model. The model shows that the local information are habituated to train the model, and the model update of new weights are gradient values are updated to the central server, aggregating the local model updates and creating a global machine learning model. This global protype is shared with all clients to use. This communica- Figure 1: Federated machine learning framework. tion of model updates from the collaborative client node to the server and the global model to the client collaborator has proven to attack where the attacker can retrieve the data from the model. This reverse data recovery by the attacking node raises the security and privacy issue.
The blockchain is a distributed ledger that keeps records like a series of blocks. The blockchain is created by the cryptographic hash of the prior block. This blockchain is immutable because the data can’t be modiifed without afecting the existing block data, which in turn cannot be altered without changing its prior block data. Every time data is added in the form of tractions, it will be updated on each and every peer node.
Figure 2 shows the blockchain-enabled federated
machine learning. The weight update from the client will
be verified by the miner using the consensus algorithm Figure 2: Blockchain-enabled Federated machine learning.
and some unique thrust test mechanisms. If the node
passes the thrust test and is only allowed to update to
the global model, the poisoning attack can be avoided leaks. Federated learning with blockchain created a
disthrough this mechanism. Adaptation of blockchain tech- tributed model for big data-driven cognitive computing
nology between the server and client node can ensure (D2C). Federated learning solves "data island" with data
sustaining information security and privacy [
There are many blockchain-based frameworks for fed- incentive mechanisms, decentralization, and poisoning erated machine learning proposed in the literature. Few resistance. Blockchain-enabled federated learning accelof the recent literature is discussed in the following para- erates member selection and verification. D2C outpergraphs. Industry 4.0 automation is advancing cogni- forms leading models in extensive review and assessment. tive computing, an AI concept miming human cognition. Industry 4.0 manufacturing performance improved by AI and machine learning technologies have enhanced federated learning model for big D2C improves perfordecision-making and data-driven intelligent manufactur- mance and privacy issues related to cognitive computing ing. The challenges, including poisoning assaults, perfor- improved significantly. mance, and limited data resources, must be handled. By integrating blockchain into federated learning, cog
Recent studies minimally examined the topic, resulting nitive computing may increase accuracy, poisoning
atin unpredictable performance, ineficiency, and privacy tack resistance, and incentive mechanism for Industry
4.0 automation. An optimization model employs a mod- permissioned blockchain architecture. Local
modificaified Markovian decision process to increase poisoning tions are encrypted using homomorphic threshold
enresistance and accuracy. The CIFAR – 10 dataset is used cryption and then added to a distributed ledger. The
for global accuracy prediction with the time setting of privacy and security flaws in the federated learning
sys4 rounds with an arrival time 0.7; the global accuracy is tem have been exposed, but the suggested approach has
0.82 [
As a cutting-edge innovation, the digital twin has gested device runs on a 2.10GHz CPU and 16GB of RAM
tremendous potential. with the fast growth of the In- running Ubuntu 20.04.1 LTS. To train machine learning
ternet of Things (IoT), which links physical components models, PyTorch is employed. Model accuracy is little
with digital space to optimize physical systems. Lim- impacted by homomorphism encryption [
Federated learning (FL) allows user data privacy- same machine learning model in parallel with no
sharpreserving machine learning (ML) model training. Exist- ing of information. Due to the hub of the framework
ing FL techniques can enable collaborative ML, but secure and the untrusty of clients, traditional federated learning
sharing of training data is dificult, especially with adver- systems are accessible to poisoning attacks from
malisarial FL clients. FL security problems and severe privacy cious clients and servers. A blockchain-based
Privacyrules require a solid and trusted FL infrastructure, such as preserving Byzantine-robust Federated Learning (PBFL)
blockchain. Scalability and client engagement issues limit strategy was proposed to reduce the influence of the
blockchain-based solutions’ industrial strength. Block malicious clients and central server. Cosine similarity
chain integration Classic FL, Hyperledger Fabric, and is employed to evaluate Clients with malicious intent
gamification are all intended to be integrated via FL. It is uploading malicious gradients, and completely
homoadvised that FL asynchronous and synchronous collabo- morphic encryption is used for safe aggregation. This
rative activities use a secure application for closing and system has a trustworthy root and uses blockchain for
signing of. The immutable ledger of an enterprise-level transparency. The scheme’s calculation in ciphertext
setblockchain network may be implemented at many FL tings is explained. The poisoning-resistant technique can
levels to improve both industrial security and auditable obtain comparable outcomes to the non-attack FedSGD
traceability. We evaluate three datasets to demonstrate scheme even with a minimal root dataset. A balanced
how the enhanced security of the FL process allows for client data distribution evaluates the technique [
For multi-party machine learning, federated learning erated deep learning-based intrusion detection system
is the answer to the problem of protecting the confiden- (FED-IDS) is proposed that ofloads server learning to
distiality of training data. There are potential security risks persed vehicle edge nodes to efectively identify threats.
associated with using this method. A trustworthy and FED-IDS uses a factors-aware transformer network to
secure federated learning platform shared by many data acquire spatial-temporal vehicle trafic flows for attack
owners is being developed using Hyperledger Fabric, a classification. Blockchain-managed federated training
3. Conclusion
lets numerous edge nodes provide distributed, safe, and
reliable training without a centralized authority. As a
safety measure, miners check distributed local updates Federated learning is a distributive machine learning
from participating vehicles before they are added to the approach that is a promising technology to uses a
masblockchain. FED-IDS outperformed state-of-the-art meth- sive amount of data to make a reliable machine learning
ods on Car-Hacking and TON IoT datasets. It validates model. Bock chain is the distributive ledger mechanism
the cyber-protection of intelligent transportation system that uses the hashing principle and consensus mechanism
networks. The proposed FED-IDS with Accuracy: 1.69%, to maintain immutable records. Since both technologies
F1- score: of 2.3% [
Blockchain-enabled Federated Learning (BFL) securely can be merged to make more advancements. This article
stores model changes on the blockchain. Mining delays reviewed such various merged schemes in various
dotraining. Mobile devices have CPU and energy limits. The mains of the field and applications. The review shows
machine learning model owner (MLMO) must choose that merging those two technologies makes the more
the energy and data used by devices for training and promising solution for their reliable deployment.
the block production rate to reduce the mining costs
and system latency while attaining the goal accuracy. Acknowledgment
Deep reinforcement learning is suggested for MLMO
decision-making under BFL uncertainty. The proposed This paper was supported by the Blockchain
TechnolDRL scheme provides up to a 12% improvement in latency ogy Laboratory at Belgrade Metropolitan University,
Belover the Greedy approach [
To train a global deep learning model from a little amount of data from several hospitals, blockchain-based federated learning is recommended. Blockchain tech- References nology verifies data, and federated learning trains the model worldwide while protecting organization privacy.
A data normalization method was proposed to handle data heterogeneity from hospitals with diverse CT scanners. Patients with COVID-19 were identified using a Capsule Network-based segmentation and classification approach. Cooperatively training a global model with federated learning and privacy using blockchain technology.
COVID-19 patient data is collected for the study. Updated
data enhances CT image recognition in the proposed
framework. Federated blockchain and capsule network
with the performance of 98.68% of improved detection of
COVID – 19 [
IIoT devices are widely used in intelligent industries.
Blockchain is strong yet susceptible to hacks. Smart
factory blockchain-based IIoT networks must detect
abnormalities to prevent attacks. Federated learning is
employed to construct block hunter, a threat-hunting
framework that automatically hunts for threats in
blockchainbased IIoT networks. Block hunter uses federated
machine learning models and a cluster-based architecture to
discover anomalies. Block hunter IIoT network federated
threat hunting approach that detects unusual activity
while protecting privacy. The block hunter can
accurately identify unusual behaviors with low bandwidth
and FedAvg approach is applied with the detection of
95% [