My research looks for new synergies between software engineering and artificial intelligence. Selected highlights:
A domain-independent mutation operator for Genetic Programming that uses Masked Language Modeling with a BERT transformer and deep reinforcement learning to suggest node replacements. Evaluated on three benchmark domains, it outperforms conventional and state-of-the-art mutation operators in convergence speed and solution quality.
A multi-parent crossover operator for genetic algorithms that uses deep reinforcement learning and an encoder-decoder architecture to learn which genes to select. It outperforms all baseline operators on two benchmark domains.
A recording of a lecture given at the international colloquium “AI meets Software Quality”. It presents ongoing work on creating a synergy between artificial intelligence and software engineering, with Behavioral Programming at its core.
Shows that specifying multiple requirements in one Petri Net obliges modelers to define how the requirements combine, which often leads to over-specification and incorrect models. Compares this with Behavioral Programming and proposes a tool for translating between the two.
A Python library for evolutionary computation, licensed under GNU GPL v3.0 and compatible with scikit-learn. It supports genetic algorithms, genetic programming, coevolution and multi-objective optimization, among other paradigms.
A paradigm for developing context-aware systems that combines Behavioral Programming with context idioms specifying when scenarios are relevant. Case studies in robotics and IoT show measurable advantages over behavioral programming alone and context-oriented programming alone.