SRM INSTITUTE OF SCIENCE AND TECHNOLOGY

Ramapuram, Chennai – 600 089

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

18CSP107L-MINOR PROJECT
Federated Learning in IOT : Privacy and Efficiency Consideration

BATCH NUMBER : 8
Team Members Supervisor
Priyanshu Dash (RA2111027020088) Dr. A. Umamageswari
Gurucharan S (RA2111027020079) Associate Professor / CSE
Mohammed Safrith D
(RA2111027020068)
 AGENDA
• Abstract
• Objective
• Scope and Motivation
• Existing System
• Literature Survey
• Summary of Issues
• Proposed System
• Advantages of Proposed System
• Novel idea
• Architecture Diagram
• Tools to be used
• Application
• Conclusion
• References (Base paper to be included)
 ABSTRACT
This project investigates the application of federated learning in
IoT environments to enhance privacy and efficiency. By keeping
data on local devices and only sharing model updates, federated
learning mitigates privacy risks and reduces communication
overhead. The goal is to develop and evaluate a federated
learning framework tailored for IoT systems.
 OBJECTIVES
• To Implement federated learning to keep data on local IoT devices.

• To Optimize communication and computational overhead and evaluate

privacy and efficiency improvements.

• To develop a functional federated learning framework for IoT and apply

the framework to a real-world IoT scenario.
 SCOPE AND MOTIVATION
SCOPE
The rapid growth of IoT devices has led to increased data
generation, raising significant privacy and efficiency concerns.

MOTIVATION

Traditional centralized learning methods are inefficient

and pose privacy risks. This project aims to leverage
federated learning to address these issues, ensuring secure
and scalable machine learning in IoT applications.
EXISTING SYSTEM
 LITERATURE SURVEY
S.N Title Of Paper Journal/ Year Inference
o Conference
1. A Survey on Federated IEEE Internet 2022 This article classifies Federated
Learning: The Journey of Things Learning topics, addressing
From Centralized to Journal technical challenges, system
Distributed On-Site models, application areas, and open
Learning and Beyond ( Volume: 8, research directions.
Issue: 7)

2. Federated learning‐based Article - 2022 This paper reviews 39 FL-based

IoT: A systematic Wiley Onlne IoT articles from 2018 to March
literature review Library 2022, highlighting key evaluation
factors like accuracy (29%) and
epoch (23%), and discussing future
research challenges.
 LITERATURE SURVEY
S.N Title Of Paper Journal/ Year Inference
o Conference
3. Decentralized Machine Journal of 2023 SL is a decentralized ML
Learning Systems system that combines edge
Architecture, computing, blockchain-based
Volume 142
P2P networking, and
coordination.
4. A Survey on Cornell 2023 This survey comprehensively
Decentralized University - summarizes existing
Federated Learning ML Article decentralized FL approaches
proposed in the
literature. Furthermore, it
identifies emerging challenges
and suggests promising
research directions in this
under-explored domain
 LITERATURE SURVEY
S.N Title Of Paper Journal/ Year Inference
o Conference
5 A Review on Federated MDPI Special 2022 Federated learning (FL) involves
Learning and Machine Issue - multiple consumers collaboratively
Learning Approaches: Foundations solving ML problems with a central
Categorization, and
Application Areas, and Challenges of collector ensuring data
Blockchain Technology Interpretable privacy.Challenges include
ML efficiency and dataset size,
prompting exploration of new FL
applications.
6. Federated Analytics With IEEE 2024 FDG trains a global model for new
Data Augmentation in Transactions clients, emphasizing privacy and
domain shifts. Current research
Domain Generalization on Machine
overlooks data augmentation and
Toward Future Networks Learning in knowledge leveraging.
Communicati Experiments reveal significant
ons and improvements: up to 5.12%
Networking accuracy boost in classification and
( Volume: 2) 0.22 increase in regression R-
squared index.
 LITERATURE SURVEY
S.N Title Of Paper Journal/ Year Inference
o Conference
7. A Survey on Federated IEEE Internet of 2021 FL is a distributed ML strategy that
Learning for Resource- Things Journal ( learns from decentralized edge
Constrained IoT Volume: 9, clients, ensuring data privacy
Issue: 1)
Devices through on-device training. This
survey examines FL's feasibility in
such environments, exploring
existing studies, implementation
challenges,etc. It addresses the
question of training distributed ML
models for IoT devices with limited
processing, bandwidth, and storage.
8. The performance 2024 - 11th 2024 FL gathers knowledge from dispersed
analysis of Federated International edge clients, enabling on-device
Conference on training and privacy preservation.
Learning Methods for However, computational assumptions
Computing for
IoT with Big Data at edge devices may not align with
Sustainable
Global reality, posing challenges for FL in
IoT networks. This study provides an
Development.
introduction to FL and examines its
implementation challenges in
 LITERATURE SURVEY
S.N Title Of Paper Journal/ Year Inference
o Conference
9. Efficient Security and IEEE Internet 2024 FL in IoT faces poisoning attacks
Privacy-Enhanced of Things where malicious devices compromise
Federated Learning Journal models with crafted gradients.This
Against Poisoning Attacks ( Volume: 11,
Issue: 8) employs cryptography to obfuscate
gradients, introducing secure
computation and verifiable protocols to
combat attacks, preserving privacy and
ensuring accuracy.

10. Hybrid IoT Device IEEE Internet 2024 FL enables collaborative model training
Selection With Knowledge of Things across IoT devices, but resource
Transfer for Federated Journal
constraints pose challenges. Our
Learning ( Volume: 11,
Issue: 7) method formulates resource constraints
as a multiobjective problem, obtaining
Pareto-optimal solutions, and utilizes
KT mechanisms to expedite
convergence.
 SUMMARY OF ISSUES
1. Data Heterogeneity
 Non-IID Data: Data across different devices may not be independent and identically distributed (IID), causing model
training difficulties.

 Variability in Data Quality: The quality and quantity of data can vary significantly between devices, affecting model
performance.

2. Communication Overhead
 Bandwidth Limitations: Frequent communication between devices and the central server can strain bandwidth and network
resources.

 Latency: High latency in communication can slow down the training process.

3. Privacy and Security

 Data Leakage: Though FL aims to preserve privacy, there is still a risk of data leakage through model updates.

 Malicious Attacks: Devices can be compromised, leading to poisoning attacks (where false data is sent) or model inversion
attacks.

4. Resource Constraints
 Computational Power: Not all devices have the same computational capabilities, which can lead to imbalances in
processing and training times.

 Battery Life: Training can be resource-intensive, draining battery-powered devices.

 PROPOSED SYSTEM

1. Device Layer
IoT Devices: Collect and process local data, train local models, and share model updates with cluster
heads.
Local Training: Perform model training on local data and apply differential privacy techniques to updates.
2. Cluster Layer
Dynamic Clustering: Form clusters based on proximity, network conditions, and device capabilities.
Cluster Heads: Aggregate updates from devices within the cluster and send aggregated updates to the
central server or intermediate aggregators.
3. Intermediate Layer
Intermediate Aggregators: Further reduce communication overhead by aggregating updates from
multiple cluster heads before sending to the central server.
4. Central Layer
Central Server: Aggregates updates from cluster heads or intermediate aggregators, updates the global
model, and disseminates it back to cluster heads.
 ADVANTAGES OF PROPOSED SYSTEM
1. Hierarchical Aggregation:
•Aggregates model updates at the cluster level before sending them to the central server, reducing data
transmission and bandwidth usage.
•Intermediate aggregators in large deployments handle some aggregation tasks, decreasing the load on the
central server.
2. Adaptive Communication Protocol:
•Bandwidth Optimization: Adjusts update frequency and size based on network conditions and device
capabilities, ensuring efficient bandwidth usage and preventing congestion.
•Latency Reduction: Prioritizes updates based on network latency and reliability for timely model
updates, accelerating the training process.
3. Dynamic Clustering:
Groups devices based on proximity and network conditions, enabling efficient local communication within
clusters and reducing data transmission distance and time.
 NOVEL IDEA
This project combines differential privacy or similar techniques
with communication efficiency measures (selective participation,
model compression) to achieve a privacy-preserving and efficient
federated learning system specifically tailored for smart home
anomaly detection.
ARCHITECTURE DIAGRAM
 TOOLS TO BE USED
• TensorFlow Federated (TFF): A library for federated learning designed to work with
TensorFlow, enabling simulations of federated algorithms on local devices.
• PySyft: An open-source library for encrypted, privacy-preserving machine learning,
supporting federated learning, differential privacy, and encrypted computation.
• PyTorch: A popular deep learning library that can be used alongside PySyft for federated
learning experiments.
• NumPy and Pandas: Essential libraries for data manipulation and numerical computations.
• Scikit-learn: A machine learning library that can be used for preliminary data analysis and
traditional machine learning algorithms.
 APPLICATION
Smart Homes: Improve privacy and efficiency in smart home devices.

Healthcare IoT: Protect patient data in medical monitoring systems.

Industrial IoT: Enhance data security and efficiency in manufacturing.

Smart Cities: Enable secure data processing in urban management systems.

Wearable Devices: Ensure privacy and efficiency in health and fitness trackers.
CONCLUSION
 REFERENCES
[1] L.S. Vailshery, "Number of IoT connected devices worldwide 2019–2021, with forecasts to 2030,"
2022.
[2] S.F. Ahmed, M.S.B. Alam, S. Afrin, S.J. Rafa, N. Rafa, and A.H. Gandomi, "Insights into Internet of
Medical Things (IoMT): Data fusion, security issues and potential solutions," Inf. Fusion, vol. 102, 2024.
[3] R. Sanchez-Iborra and A.F. Skarmeta, "TinyML-enabled frugal smart objects: Challenges and
opportunities," IEEE Circuits Syst. Mag., vol. 20, no. 3, pp. 4-18, 2020.
[4] P.P. Ray, "A review on TinyML: State-of-the-art and prospects," J. King Saud Univ. - Comput. Inf.
Sci., vol. 34, no. 4, pp. 1595-1623, 2022.
[5] Q. Li, Z. Wen, Z. Wu, S. Hu, N. Wang, X. Liu, and B. He, "A survey on federated learning systems:
Vision, hype and reality for data privacy and protection," IEEE Trans. Knowl. Data Eng., vol. 35, pp.
3347-3366, 2023.
[6] M. Kachuee, S. Fazeli, and M. Sarrafzadeh, "ECG heartbeat classification: A deep transferable
representation," presented at the IEEE International Conference on Healthcare Informatics, 2018, pp. 443-
444.
[7] R.F. Vitor, "Car trips data log," 2017. [Online]. Available:
https://www.kaggle.com/datasets/vitorrf/cartripsdatamining
[8] V. Tsoukas, E. Boumpa, G. Giannakas, and A. Kakarountas, "A review of machine learning and
TinyML in healthcare," in PCI ’21: Proc. 25th Pan-Hellenic Conference on Informatics, ACM, 2022, pp.
69-73.
[9] S.K. Lo, C.W. Q. Lu, H.-Y. Paik, and L. Zhu, "A systematic literature review on federated machine
learning: From a software engineering perspective," ACM Comput. Surv., vol. 54, no. 5, pp. 1-39, 2021.
[10] H. Ren, D. Anicic, and T. Runkler, "TinyOL: TinyML with online-learning on microcontrollers,"
presented at the International Joint Conference on Neural Networks (IJCNN), IEEE, 2021, pp. 1-8.
[11] C.R. Banbury, V.J. Reddi, M. Lam, et al., "Benchmarking tinyml systems: Challenges and
direction," 2020.
[12] R. David, J. Duke, A. Jain, et al., "TensorFlow Lite Micro: Embedded Machine Learning on
TinyML Systems," in Proceedings of the 4th Machine Learning and Systems (MLSys 2021), 2021.
[13] L. Lai, N. Suda, and V. Chandra, "CMSIS-NN: Efficient neural network kernels for arm Cortex-M
CPUs," 2018.
[14] Apache TVM Project, "https://tvm.apache.org/", 2021.
[15] L. Ravaglia, M. Rusci, D. Nadalini, A. Capotondi, F. Conti, and L. Benini, "A tinyml platform for
on-device continual learning with quantized latent replays," IEEE J. Emerg. Sel. Top. Circuits Syst.,
vol. 11, no. 4, pp. 789-802, 2021.
[16] J. Lin, W.-M. Chen, Y. Lin, J. Cohn, C. Gan, and S. Han, "Mcunet: Tiny deep learning on iot
devices," Adv. Neural Inf. Process. Syst., vol. 33, pp. 11711-11722, 2020.
[17] J. Montiel, M. Halford, S.M. Mastelini, G. Bolmier, R. Sourty, R. Vaysse, A. Zouitine, H.M.
Gomes, J. Read, T. Abdessalem, and A. Bifet, "River: machine learning for streaming data in Python,"
2020.
[18] K. Kopparapu, E. Lin, J.G. Breslin, and B. Sudharsan, "TinyFedTL: Federated transfer learning on
ubiquitous tiny IoT devices," in IEEE Int. Conf. on Pervasive Computing and Communications
Workshops and Other Affiliated Events, IEEE, 2022, pp. 79-81.
[19] M.M. Grau, R.P. Centelles, and F. Freitag, "On-device training of machine learning models on
microcontrollers with a look at federated learning," in Conf. on Information Technology for Social
Good, 2021, pp. 198-203.
[20] Y.D. Kwon, R. Li, S.I. Venieris, J. Chauhan, N.D. Lane, and C. Mascolo, "TinyTrain: Deep neural
network training at the extreme edge," 2023.