Abstract
Intrusion Detection Systems (IDS) have an increasingly important role inpreventing exploitation of network vulnerabilities by malicious actors. Recentdeep learning based developments have resulted in significant improvements inthe performance of IDS systems. In this paper, we present FetFIDS, where weexplore the employment of feature embedding instead of positional embedding toimprove intrusion detection performance of a transformer based deep learningsystem. Our model is developed with the aim of deployments in edge learningscenarios, where federated learning over multiple communication rounds canensure both privacy and localized performance improvements. FetFIDS outperformsmultiple state-of-the-art intrusion detection systems in a federatedenvironment and demonstrates a high degree of suitability to federatedlearning. The code for this work can be found athttps://github.com/ghosh64/fetfids.