AI for Security and Security of AI

Artificial Intelligence (AI) constitutes an umbrella of techniques, and has proven to provide flexible, adaptable solutions to wide variety of security solutions. These techniques typically include Neural Networks, Fuzzy Logic, Evolutionary Computation, Data Mining, Cellular Automata, Immunological Computation, Game Theory, and other computational intelligence models. Over the last 30 years, AI-based approaches were used to build tools for real-tine monitoring, malware detection, log analysis, intrusion detection, etc., providing cross-linking solutions to different cyber security applications. This talk will cover cyber-attack landscapes and how different AI-guided strategies have been used for defenses. The effectiveness and limitations of pre-trained Generic LLMs in identifying and mitigating emerging cyber threats. 

For securing AI models against adversarial attacks, I will discuss our research on dual-filtering (DF) strategy that could mitigate input data and model manipulations for wide-range of adversarial attacks. The output decision boundary inspection using a classification technique automatically reaffirms the reliability and increases the trustworthiness of any ML-Based decision support system. Unlike other defense strategies, our DF defense technique does not require adversarial sample generation and regular updating of decision boundary for detection makes the defense system robust to adaptive attacks. 

The talk will cover how integrated AI techniques can bolster holistic cyber defense and why it is also very important to ensure secure and trustworthy AI systems for safety-critical applications, 

References:

● Machine learning in cybersecurity: a comprehensive survey. Dasgupta, D., Akhtar, Z. and Sen, S. The Journal of Defense Modeling and Simulation, 19(1), pp.57-106, 2022.

● AI-Powered Ransomware Detection Framework by Subash Poudyal and Dipankar Dasgupta. IEEE Symposium Series on Computational Intelligence (SSCI), pp. 1154-1161, Australia, 2020.

● Dual-Filtering (DF) Schemes for Learning Systems to prevent Adversarial Attacks, Dasgupta, D., Gupta, K.D. Complex Intell. Syst. 9, 3717–3738 (2023). https://doi.org/10.1007/s40747-022-00649-1.

● AI vs. AI: Viewpoints. D. Dasgupta, Technical Report No.(CS-19-001, The University of Memphis). An Invited talk by Dr. Dasgupta on IEEE DAY 2022 as an IEEE Distinguished Lecturer (https://www.youtube.com/watch?v=EmMXkwJdgAo) .

● Adversarial Attacks and Defense. A tutorial by D. Dasgupta and Kishor Datta Gupta. https://www.slideshare.net/slideshow/adversarial-attacks-and-defense/250801927.