public.loading

18th Nov 26

Talk Ideas with Tiago Rolo and Manuel Drago

Speaker: Tiago Rolo

Date: 25th of November, 2026

Time: 4:00pm

Place: Room E5.6 (CSLab Room) and Online

Presentation title: “Should Agents Do Their Own Concurrency Control”

Short bio:

Tiago Rolo completed a Bachelor’s in Informatics at the University of the Azores and a Master’s in Informatics Engineering at the University of Coimbra, where he is now a PhD Student and a member of the SSE group at CISUC. His research focuses on distributed systems protocols for LLM-based multi-agent systems, studying how shared logs can help agents act consistently on shared data.

Abstract:

Shared logs have recently been adopted for AI agents, but current systems do not address concurrency control when agents mutate shared data. Each agent interprets the log on its own, and application invariants can silently break when an agent acts on state that changed while it was still reasoning, or hallucinates an operation that is invalid in any order. In this talk, we ask whether an oracle that handles agent operations in the log can avoid these violations. Applications choose how much the oracle does by assigning a type to each shared object, so the oracle orders every operation (Total Order), orders only those that might break invariants (RedBlue), or also rejects operations that would exceed a pre-established bound (Bounded Counter). We test this on scheduling and logistics tasks, where agents either coordinate through the log alone or act through the oracle under each object type. Beyond whether the agents complete their tasks, we measure how often invariants break, how many operations are stale or hallucinated, and how many tokens each setup consumes.

_____________________________

Speaker: Manuel Drago

Date: 25th of November, 2026

Time: 4:00pm

Place: Room E5.6 (CSLab Room) and Online

Presentation title: “Standard Countermeasure Selection in Standard-Compliant Cybersecurity Risk Assessment”

Short bio:

Manuel Drago received his Bachelor's degree in Computer Science and his Master's degree in Resilient and Secure Cyber-Physical Systems from the University of Florence, Italy. He is currently pursuing a Ph.D. in Computer Science at the same university, where he is also a member of the Resilient Computing Lab.  His research focuses on the dependability and security of cyber-physical systems, with particular interests in cybersecurity risk assessment, stochastic modeling, and quantitative evaluation methodologies. His work involves the development and application of model-based approaches for analyzing security risks, assessing system resilience, and supporting decision-making in safety-critical environments.

Abstract:

Cybersecurity risk assessment standards provide structured methodologies for identifying threats, evaluating risks, and determining whether residual risk is acceptable. However, when additional security measures are required, these standards offer limited guidance on how to select an appropriate combination of countermeasures while considering their effectiveness and implementation costs.  This talk explores how mathematical optimization can support countermeasure selection within standard-compliant cybersecurity risk assessment processes. The proposed methodology combines attack graphs, which represent possible multi-stage attack scenarios, with quantitative risk evaluation and optimization techniques to identify suitable mitigation strategies.  The approach considers two complementary decision-making perspectives: selecting the minimum-cost set of countermeasures needed to achieve an acceptable residual risk level, and identifying the most effective risk reduction achievable within a given budget. Particular attention is given to how these optimization techniques can be integrated into existing risk assessment workflows, supporting more systematic, transparent, and quantitatively justified cybersecurity decisions.

This website uses cookies to improve your experience. Read More