Područje interesa Odjela za računalnu inteligenciju je teorija, oblikovanje, primjena i razvoj biološki i lingvistički motiviranih računalnih paradigmi s naglaskom na neuronske mreže, konekcijske sustave, genetičke algoritme, evolucijsko programiranje, neizrazite sustave i hibridne inteligentne sustave koji su temeljeni na ovim paradigmama.
Odjel za računalnu inteligenciju Hrvatske sekcije IEEE poziva Vas na predavanje
"Metaheuristic Attacks on Verifiable AI Inference"
koje će održati doc. dr. sc. Luca Mariot sa Sveučilišta u Twenteu, u petak, 30. listopada 2026. godine, s početkom u 14.15 sati u sivoj vijećnici Fakulteta elektrotehnike i računarstva Sveučilišta u Zagrebu, Unska 3.
Predavanje će se održati na engleskom jeziku, a predviđeno trajanje s raspravom je 60 minuta. Predavanje je otvoreno za sve zainteresirane, a posebno pozivamo studente.
Više informacija o predavanju i predavaču pročitajte u opširnijem sadržaju obavijesti.
Sažetak predavanja:
When large neural network models run on the cloud, clients have no guarantee that the answers they receive were computed by the original model. Zero-knowledge proofs can provide this guarantee, but they remain far too slow for large models. Recent statistical proofs of inference avoid this cost by checking only a few randomly sampled paths through the network’s committed activation trace. This makes verification fast, but if an adversary concentrates the attack on neurons that the verifier rarely visits, it may go undetected. In this talk, we formulate the attack as an optimization problem on the network graph: find a set of neurons to corrupt that are unlikely to be checked, while still causing significant damage to the output. We then suggest using Ant Colony Optimization (ACO) to address this optimization problem, especially against non-uniform and relevance-guided sampling strategies, and outline open questions on co-evolving attackers and verifiers.
Životopis predavača:
Luca Mariot is an assistant professor at the University of Twente, The Netherlands. His main research interests lie at the intersection of cryptography and artificial intelligence, specifically focusing on nature-inspired computational models and optimization techniques to design cryptographic primitives. Previously, Luca was a postdoc researcher at Radboud University and TU Delft, the Netherlands, and at the University of Milano-Bicocca, Italy. He received his PhD in Computer Science under a double degree agreement, from the University of Milano-Bicocca and the Université Côte d'Azur, France. Up to now, Luca published more than 70 peer-reviewed papers in various journals and conferences related to cryptography and natural computing.