Scribd
Upload
5 views1 page

Federated Learning for Cybersecurity Privacy

The document outlines the challenges faced by traditional centralized cybersecurity systems, particularly regarding data privacy and vulnerability to attacks. It introduces Federated Learning (FL) as a solution that allows collaborative model training without sharing raw data, but highlights the risks of gradient inversion attacks. The proposed thesis aims to develop a comprehensive FL framework that integrates Homomorphic Encryption and Secure Aggregation to enhance privacy, efficiency, and detection accuracy in cyber threat detection.

Uploaded by

navneet
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
5 views1 page

Federated Learning for Cybersecurity Privacy

The document outlines the challenges faced by traditional centralized cybersecurity systems, particularly regarding data privacy and vulnerability to attacks. It introduces Federated Learning (FL) as a solution that allows collaborative model training without sharing raw data, but highlights the risks of gradient inversion attacks. The proposed thesis aims to develop a comprehensive FL framework that integrates Homomorphic Encryption and Secure Aggregation to enhance privacy, efficiency, and detection accuracy in cyber threat detection.

Uploaded by

navneet
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Navneet Mishra

MTech CSE
250036

Problem Definition & Research Gap

Cybersecurity systems rely heavily on analysing vast amounts of sensitive data, such as
network traffic logs, intrusion records, and malware traces. Traditional centralized
approaches for cyber threat detection face two major challenges: (i) organizations are often
reluctant to share raw data due to privacy and confidentiality concerns, and (ii) centralized
data storage introduces single points of failure and increases the attack surface.

Federated Learning (FL) has emerged as a promising paradigm that allows multiple
participants to collaboratively train a global model without sharing raw data. Instead,
participants only exchange model updates or gradients. However, recent studies (e.g., Deep
Leakage from Gradients (DLG) and improved DLG (iDLG) have demonstrated that gradients
and model updates are vulnerable to gradient inversion attacks, where adversaries can
reconstruct private training data or infer sensitive labels. This presents a severe privacy risk
when FL is applied to the cybersecurity domain, where leaked information may reveal
proprietary defence mechanisms, system vulnerabilities, or sensitive user activity logs.

Existing work in FL security has primarily focused on either accuracy improvement or


applying single privacy-preserving techniques such as Homomorphic Encryption (HE) or
Secure Aggregation (SA). However, there remains a critical research gap in developing a
comprehensive and practical privacy-preserving FL framework for cyber threat detection
that simultaneously ensures:

• Robust privacy against gradient leakage and model inversion attacks,

• Efficient aggregation of updates from multiple participants, and

• High detection accuracy despite the added privacy-preserving mechanisms.

My thesis addresses the gap by proposing a novel FL framework for cyber threat detection
that integrates Homomorphic Encryption (HE) and Secure Aggregation (SA). The proposed
framework aims to protect against gradient leakage while enabling collaborative learning
across distributed cybersecurity participants without compromising detection accuracy.

You might also like