Key Publications

2019: "RelExt: Relation Extraction using Deep Learning Approaches for Cybersecurity Knowledge Graph Improvement " accepted at ASONAM 2019
We extracted word embeddings of cybersecurity text from open source. We used a deep neural network to find how pairs of cyber entities are related. The new relationships we found for pairs of cyber entities aided us to form semantic triples which could easily be asserted into a cyber-knowledge graph. Tags: NLP, AI, Cybersecurity Knowledge Graphs

2019: "Preventing poisoning attacks on AI based Threat Intelligence Systems" accepted at IEEE MLSP 2019
We used semi-supervised learning to identify poisoning attacks which could harm threat intelligence extractors. We classified open source data like reddit posts about cyber entities.  Tags: NLP, AI, Semi-supervised learning

2020: "NAttack! Adversarial Attacks to bypass a GAN based classifier trained to detect Network Intrusion" accepted at IEEE Big Data Security 2020

We designed a GAN based classifier which improves the accuracy to detect suspicious network activity. We trained our classifier on a pretty big dataset. The GAN based classifier relaxed the class boundaries which should have made this immune to adversarial attacks. However, we proved that the GAN based classifier is susceptible to adversarial attacks, after a certain number of iterations.

Tags: Generative Adversarial Networks, AI, Adversarial Attacks,Network security

2020: "A Smart-Farming Ontology for Attribute Based Access Control" accepted at IEEE Big Data Security 2020

We created a novel ontology to represent the data coming from smart farms. This could be extended to other CPS systems

Tags: Smart Security, Knowledge Graphs

2020: "Using Knowledge Graphs and Reinforcement Learning for Malware Analysis" accepted at IEEE Big Data Security 2020

We combined knowledge graphs and used the prior knowledge to influence  RL algorithm's  parameters for malware analysis

Tags: Reinforcement Learning, Knowledge Graphs

2022: "Knowledge guided Two Player Reinforcement Learning for Cyber Attacks and Defenses" accepted at IEEE ICMLA 2022

We combined knowledge graphs and used the prior knowledge to influence  RL algorithm's  parameters for simulated cyber-attacks and defenses. We investigated the effect of knowledge influence

Tags: Reinforcement Learning, Knowledge Graphs

2021: "Generating Fake Cyber Threat Intelligence Using Transformer-Based Models" accepted at IEEE IJCNN 2021

We explored if we can create fake cyber threat intelligence through GPT.  Turns out  they are good enough to fool experts.

Tags: LLM, Generative Models,  Knowledge Graphs

2023: "Offline RL+CKG: A hybrid AI model for cybersecurity tasks" accepted at AAAI MAKE 2023

We explored if we can make conservative Q-Learning, not so conservative with knowledge graphs and guidance

Tags: Offline Reinforcement Learning,  Knowledge Graphs