Location
https://ccse.kennesaw.edu/computing-showcase/cday-programs/spring2021program.php
Event Website
1. The project is based on applying a Federated Learning concepts towards detecting network threats for IoT devices.2. This project is by Osama Shahid with advising from Dr. Seyedamin Pouriyeh.
Document Type
Event
Start Date
26-4-2021 5:00 PM
Description
There are abundant number of IoT devices that are connected on over multiple networks. These devices can be exposed to multiple different types of network threats. Though, these devices do have security and software that does act as a wall of protection we purpose a Federated Learning (FL) approach that would allow detection of threats of a network for IoT devices. Federated Learning can be best described as decentralized training. Adhering to the GDPR rules that prevent data from being distributed, FL addressed the challenge by bringing the ML model to the data rather than the traditional method where the data had to be extrapolated and taken to the ML model. This type of setting is ideal for IoT devices (Client) that are connected to the network and can download the FL model that would allow them to keep their devices more secure. FL is different from on-device training. Once the Client(s) download the model and train on their individual local data, the updated model is shared on a central server. The central server as all the individual models shared by each client on the federated network. The central server aggregates all these models as one new global model. For our project we believe this is an ideal setting for IoT devices that are susceptible to network threats. In our Federated Network we have four clients, and we create a FL framework that allows each client to train a FL model on their local data. The model(s) are then aggregated to create a new global model. It is worth noting that each client has its’ own distinct type of threat. So, when all the models are aggregated, they have the knowledge of each of these individual models and this new model is capable to testing and detecting threats that are posed across all the devices.Advisors(s): Dr. Seyedamin PouriyehTopic(s): Security
Included in
GR-44 An efficient intrusion detection framework based on federated learning for IoT networks.
https://ccse.kennesaw.edu/computing-showcase/cday-programs/spring2021program.php
There are abundant number of IoT devices that are connected on over multiple networks. These devices can be exposed to multiple different types of network threats. Though, these devices do have security and software that does act as a wall of protection we purpose a Federated Learning (FL) approach that would allow detection of threats of a network for IoT devices. Federated Learning can be best described as decentralized training. Adhering to the GDPR rules that prevent data from being distributed, FL addressed the challenge by bringing the ML model to the data rather than the traditional method where the data had to be extrapolated and taken to the ML model. This type of setting is ideal for IoT devices (Client) that are connected to the network and can download the FL model that would allow them to keep their devices more secure. FL is different from on-device training. Once the Client(s) download the model and train on their individual local data, the updated model is shared on a central server. The central server as all the individual models shared by each client on the federated network. The central server aggregates all these models as one new global model. For our project we believe this is an ideal setting for IoT devices that are susceptible to network threats. In our Federated Network we have four clients, and we create a FL framework that allows each client to train a FL model on their local data. The model(s) are then aggregated to create a new global model. It is worth noting that each client has its’ own distinct type of threat. So, when all the models are aggregated, they have the knowledge of each of these individual models and this new model is capable to testing and detecting threats that are posed across all the devices.Advisors(s): Dr. Seyedamin PouriyehTopic(s): Security
https://digitalcommons.kennesaw.edu/cday/spring/graduateresearch/8