Sophisticated cyber attacks increasingly threaten critical infrastructures, highlighting the need for more reliable user authentication. This work proposes a deep learning–based framework that employs keystroke dynamics and Siamese neural networks to distinguish legitimate users from impostors. To address variability in password lengths, we introduce interpolation-based data fusion strategies that standardize keystroke features across datasets. The proposed framework incorporates adaptive threshold strategies to determine decision boundaries relative to a user’s baseline typing behaviour. Experimental evaluation on fused CMU and KeyRecs datasets achieved equal error rates as low as 0.11–0.12. The results demonstrate the robustness and scalability of the approach for detecting insider threats.

Static and dynamic malware analysis are widely used in cybersecurity, though attackers have adapted to these methods. To improve detection, researchers increasingly combine these methods with machine/deep learning for faster, more efficient malware classification. However, adversaries exploit weaknesses to craft adversarial malware that evades detection. We aim to design a deep learning–based C2 framework to enhance red team training and improve anomaly detection beyond reliance on automated tools.

Šiame darbe siūloma metodika, skirta gerybinės ir kenkėjiškos programinės įrangos kategorizavimui, siekiant padidinti kenkėjiškų programų aptikimo tikslumą. Metodika remiasi statinės analizės duomenimis, derinamais su šiuolaikiniais duomenų apdorojimo ir vizualizacijos metodais.