 | 2026 14-th International Scientific Conference COMPUTER SCIENCE |  |
|---|
Invited paper "Agentic AI for Cyber-Resilient Digital Twins: Trusted Autonomous Control of Intelligent Critical Infrastructures" by Assoc. Prof. Dmytro Prokopovych-Tkachenko, University of Customs and Finance, Dnipro, Ukraine; Institute of Information, Security and Law of the National Academy of Legal Sciences of Ukraine, Ukraine Abstract: The convergence of Agentic AI, digital twins, cyber-physical systems, and autonomous decision-making is changing how critical infrastructures are monitored and controlled. Yet greater autonomy also increases exposure to poisoned telemetry, malicious tool use, cascading failures, opaque reasoning, and loss of human oversight. This plenary talk presents a verification-gated framework for cyber-resilient digital twins that treats autonomy as a constrained operational capability rather than an unrestricted controller. The framework combines heterogeneous sensing, provenance- and uncertainty-aware state estimation, multi-agent planning, policy-as-code, twin-based simulation, least-privilege execution, and auditable recovery. Intelligent agents observe infrastructure states, propose defensive and recovery actions, and coordinate response; a separate verification layer checks identity, authority, safety invariants, operational impact, and confidence before any action is approved, rejected, or escalated to a human. The synthesis builds on recent work in environmental digital twins, privacy-preserving data governance, blockchain-supported evidence integrity, and uncertainty-aware passive Wi-Fi localization for critical-infrastructure sensor networks. Published component-level results show how secure engineering, calibrated uncertainty, and trusted provenance can improve the fidelity and auditability of digital-twin operations, while an integrated evaluation agenda is proposed for measuring policy-violation rates, unsafe-action rejection, service continuity, recovery time, and trace completeness. Agentic AI can support controlled autonomous or semi-autonomous cyber recovery only when verification, reversibility, uncertainty, and human governance are architectural properties of the system.
Invited paper "Grid-Connected Inverters Induced Power Surges in Electric Power System" by Prof. Sasa Sladic, University of Rijeka, Rijeka, Croatia Abstract: Recent statistics on blackouts could appear both in research papers and in media. Blackouts are in increase worldwide. Importance of this subject sparks the global interest. That is because main problem of modern electric power system is how to incorporate inconsistent power sources (especially PV and WECS). Power surges which appear and have not been always noticed could influence the temporary power and frequency of electric power system by inducing oscillations and result in chain reaction similar to those prior the Iberian blackout. This important subject has been analyzed through simulations confirming the previously published hypothesis in many different cases worldwide.
Invited paper "From Stakeholder Needs to Deployed Practice: Designing and Evaluating a Governed Generative-AI Assistant for Entrepreneurship Education" by Assoc. Prof. Angel Marinov, Technical University of Varna, Varna, Bulgaria Abstract: Universities face growing demand for individualised entrepreneurship support that human mentoring cannot scale, while students already use generative AI daily—86% in our survey—mostly without formal preparation. This talk presents the complete design chain of an institutionally governed, retrieval-augmented (RAG) assistant developed within the Erasmus+ project SEEN: a four-country stakeholder needs analysis (n = 400), the requirements derived from it, the comparison of three implementation approaches, corpus governance, and deployment on university infrastructure. It then reports the evidence from two independent evaluation stages—an expert test (18 raters, 50 scenarios, 350 scenario–criterion combinations) and a field test with 114 users—showing adequate content quality (expert mean 3.37/5), high usability, and a consistent weakness in pitch evaluation that points to the retrieval corpus rather than the model. The talk closes with practical recommendations for institutions adopting generative AI: evaluate functions rather than averages, govern the corpus as part of the pedagogy, and keep the deployment under institutional control.
Last changed on 30.08.2026, 12:18:34