What we are looking for
We are seeking innovative research proposals that explore issues at the frontier of machine learning and cybersecurity including both more secure/robust ML methods and/or ML methods to enhance cybersecurity.
The context
We are seeking innovative research proposals that explore issues at the frontier of machine learning and cybersecurity including both more secure/robust ML methods and/or ML methods to enhance cybersecurity.
AI and Machine Learning have had tremendous successes in many application domains including cybersecurity. Yet, as reflected in the recent EU AI Act, AI and ML bring along novel risks, e.g. in relation to security. This has led to the emergent field of Adversarial Machine Learning
The problem to address
David Rios Insua (ICMAT) and David Arroyo (ITEFI) are joining efforts to address core issues in relation to ML and cybersecurity. These include a novel Bayesian general framework for adversarial machine learning, a novel framework for cybersecurity risk management in systems with AI components and the application of AI in the extraction of actionable intelligence from security traits, logs and activity in the clear and the deep web. They are also exploring security issues in LLMs. The above problems are of interest for various companies that collaborate with us (Aeroengy, TheBasement, Enzyme, Unisys…) and various international collaborators (CNR-IMATI, GWU, Aalto…)
Objectives
- To investigate broader frameworks that enhance security/robustness of ML (supervised, unsupervised, reinforced)
- To introduce broader cybersecurity risk management frameworks in systems with AI components.
- To analyse connections between cybersecurity risk analysis and threat modeling.
- To improve current techniques to extract actionable intelligence from indicators of compromise and other text-based intel.
- To explore previous ideas in relation to LLMs.
Expected Outcomes
- A new general framework for adversarial machine learning.
- A novel framework for AI cybersecurity risk management.
- Securised and robustified LLMs.
- An actionable cyber threat intelligence discovery system.
- Contributions to the body of knowledge at the intersection of adversarial machine learning and cybersecurity.