CBChristina Berghegger

Christina Berghegger

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Christina Berghegger has a total of 7 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Study on Search Heuristics in Cartesian Genetic Programming

    by Christina Berghegger, Sylvain Cussat-Blanc, David Simoncini, Yuri Lavinas

    The effectiveness of search heuristics in Genetic Programming is affected by problem specific fitness landscape characteristics. The standard search strategy in Cartesian Genetic Programming is a hill climber 1+λ Evolutionary Algorithm. Besides this implementation, the field currently lacks research on alternative search heuristics, especially those designed to traverse more efficiently fitness landscapes with difficult features, such as ruggedness or neutrality. We analyze the search behavior of the classic 1+λ Evolutionary Algorithm and show that it suffers from premature convergence due to fitness landscape characteristics decreasing evolvability. Under the hypothesis that alternative Evolutionary Algorithms with higher exploration capabilities improve evolvability and performance, we compare and assess different search heuristics, notably based on Simulated Annealing and Monte Carlo Tree Search, on Multimodal Adaptive Graph Evolution, a multi-type variant of Cartesian Genetic Programming. We show that all alternative Evolutionary Algorithms increase performance on a program synthesis benchmark, challenging the standard hill climber implementation.

  • Using Monte Carlo Tree Search to Enhance Search Space Exploration in Cartesian Genetic Programming

    by Christina Berghegger, Camilo De La Torre, Sylvain Cussat-Blanc, Yuri Lavinas, David Simoncini

  • Open Access

    Analyzing Interpretable Visual Control Policy Search and Synthesis

    by Christina Berghegger, Camilo De La Torre, Giorgia Nadizar, Yuri Lavinas, David Simoncini, Hervé Luga, Dennis G. Wilson, Sylvain Cussat-Blanc

    Vision-based decision-making is relevant to many domains, including safety-critical ones where transparency matters as much as performance. Therefore, automating sequential decision making in such settings requires approaches that balance effectiveness with interpretability. While deep reinforcement learning techniques based on artificial neural networks have achieved strong performance, their black-box nature typically necessitates post hoc explainability analyses. To address this limitation, we propose an approach based on Graph-based Genetic Programming (GGP) that generates policies in the form of computer code, which is fully observable and inherently interpretable. To improve both performance and robustness, we expose GGP-based visual control policies to multiple representative conditions during optimization, mitigating convergence to strong local optima and fragility under changing conditions. Finally, to gain insights into the evolutionary search dynamics, we employ Search Trajectory Networks, an analytical and visualization tool for studying optimization behavior. Our results demonstrate that the resulting policies approach human-level performance and empirically confirm the presence of strong local optima acting as attractors during evolution, providing new insights into the behavior and potential limitations of interpretable evolutionary policy search approaches.