Scribd
Upload
5 views3 pages

Anomaly Detection in Federated Learning

The document outlines a framework for anomaly detection using a global federated CAE model, employing MSE thresholding and a sliding window majority vote to reduce false positives. It details a hierarchical federated learning approach involving local device training, regional aggregation with an adaptive heuristic function, and global blockchain integration for model integrity. Additionally, it discusses potential blockchain implementations, including Hyperledger Fabric and Ethereum, for secure model exchange and device authentication, along with proposed smart contract functions for device registration and model submission.

Uploaded by

2453005
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
5 views3 pages

Anomaly Detection in Federated Learning

The document outlines a framework for anomaly detection using a global federated CAE model, employing MSE thresholding and a sliding window majority vote to reduce false positives. It details a hierarchical federated learning approach involving local device training, regional aggregation with an adaptive heuristic function, and global blockchain integration for model integrity. Additionally, it discusses potential blockchain implementations, including Hyperledger Fabric and Ethereum, for secure model exchange and device authentication, along with proposed smart contract functions for device registration and model submission.

Uploaded by

2453005
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd

1.

anamoly detection

 Deploy the final global federated CAE model back to devices.

 Detect anomalies using Reconstruction Error (MSE) thresholding:

Anomaly if MSE>μ+3σ\text{Anomaly if } MSE > \mu + 3\sigmaAnomaly if MSE>μ+3σ

 Use a sliding window majority vote approach to minimize false positives.

2. Hierarchical Federated Learning

 Step 1: Local devices train and send compressed model updates to the regional
aggregator.

 Step 2: Regional aggregator aggregates local updates using an Adaptive Heuristic


Aggregation Function:

Wagg=αiWlocal+(1−αi)∑j∈LinjmiWjW_{agg} = \alpha_i W_{local} + (1 - \alpha_i)\sum_{j \in


L_i}\frac{n_j}{m_i}W_{j}Wagg=αiWlocal+(1−αi)j∈Li∑minjWj

where αi\alpha_iαi is the adaptive coefficient.

 Step 3: The global blockchain aggregates regional updates and records model hashes for
integrity verification.

BLOCKCHAIN IMPLEMENTATION

Anomaly if MSE>μ+3σ
🔹 Summary

Aspect Base Paper Implementation

Blockchain Role Theoretical layer for model exchange, authentication, and integrity.

Model/Platform
None (assumed “permissioned blockchain service”).
Used

Consensus
Not implemented — described conceptually (could be PBFT or Proof-of-Authority).
Mechanism

Smart Contracts Mentioned as a future direction, not implemented.


Aspect Base Paper Implementation

Future Work Implement and evaluate blockchain overhead, latency, and scalability.

Reference model

Blockchain Model Type Why Use It Tools

Supports identity management (for


Hyperledger Docker + Fabric SDK
Permissioned IoT device authentication) and
Fabric (Python/NodeJS)
efficient consensus (PBFT).

To implement smart contracts for


Ethereum (Private
Public/Private model updates; flexible integration Ganache + Truffle
Network)
with FL via [Link].

High throughput, lightweight


Tendermint / Tendermint node +
Permissioned consensus (BFT), suitable for edge
Cosmos SDK Python RPC
environments.

Quorum
For enterprise-grade permissioned
(Enterprise Permissioned Quorum + web3
blockchain with privacy.
Ethereum)

Blockchain Integration Model for HBFL (Hierarchical BFL)

Smart contract / chaincode functions (pseudocode):

function registerDevice(deviceID, pubKey, cert) {


require(![Link](deviceID))

DeviceRegistry[deviceID] = {pubKey, cert, trustScore: default}

function submitModel(deviceID, CID, round, metrics, commitmentCoeff, sig) {

require([Link](deviceID))

require(verifySignature([Link], sig, CID||round||metrics))

ModelEntryID = hash(deviceID||CID||round||timestamp)

ModelEntries[ModelEntryID] = {deviceID,CID,round,metrics,commitmentCoeff}

emit NewModel(ModelEntryID)

function submitRegionalModel(regionID, aggCID, round, contributors[], validatorSigs) {

require(verifyValidators(validatorSigs))

RegionalEntries[regionID||round] = {aggCID,contributors}

emit NewRegionalModel(regionID, aggCID)

function updateTrust(deviceID, delta) onlyValidator {

DeviceRegistry[deviceID].trustScore += delta

You might also like