Piotr Duda

This study addresses the vulnerability of AI systems to adversarial attacks by extending the DeepFool algorithm. The paper proposes four new approaches and evaluates them according to a set of criteria. The methods are inspired by various optimisation algorithms. One of the proposed improvements adds the independent refinement stage, which reduces the final perturbation without extra gradient computations. Experimental results show that the appropriately modified algorithm reaches the decision boundary in fewer steps and with fewer gradient evaluations, while the refinement stage further decreases the magnitude of the perturbation. The combined approach can improve attack efficiency and reduce detectability, suggesting the potential for a wider application of advanced optimisation techniques in adversarial example generation.

Socio-cognitive computing is a paradigm developed for the last several years in our research group. It consists of introducing mechanisms inspired by inter-individual learning and cognition into metaheuristics. Different versions of the paradigm have been successfully applied in hybridizing Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), Genetic Algorithms, Differential Evolution, and Evolutionary Multi-agent System (EMAS) metaheuristics. In this paper, we have followed our previous experiences in order to propose a novel mutation based on sociocognitive mechanism and test it based on Evolution Strategy (ES). The newly constructed versions were applied to popular benchmarks and compared with their reference versions.

Sciendo provides publishing services and solutions to academic and professional organizations and individual authors. We publish journals, books, conference proceedings and a variety of other publications.