Project title An AI agent for practical derivative-free global optimization with classical and quantum perspectives

Project number MPK-49

Project implementation period 2026-09-01 – 2029-08-31

Project leader Dr. Linas Stripinis

Project objective - to develop an artificial intelligence (AI)-based optimization agent capable of adaptively selecting suitable optimization algorithms and constructing effective algorithmic combinations for solving derivative-free global optimization (DFGO) problems.

Project summary – The project aims to develop a practical Artificial Intelligence (AI)-based decision-support framework for a specific class of complex optimization problems: single-objective, continuous derivative-free global optimization. The proposed AI agent will be built upon a portfolio of classical and hybrid optimization methods, including systematically designed novel hybrid algorithms derived from underexplored algorithmic compositions and hybridization frameworks. The framework will also incorporate intelligent optimization tools such as algorithm selectors and automatic algorithm designers, enabling adaptive selection and configuration of optimization strategies depending on the characteristics of the problem instance. Selective Machine Learning (ML) techniques will be investigated to support solver selection and instance-level decisions in hybrid algorithm composition, particularly in cases where data-driven approaches can provide measurable improvements in robustness, efficiency, or computational cost. Rather than relying solely on ML, the project adopts a balanced approach that integrates algorithmic knowledge with targeted data-driven components. A benchmark-driven methodology underpins the framework, emphasizing reproducibility, transparent evaluation, and practitioner-oriented decision support. Standardized benchmark suites, open software implementations, and systematic performance analysis will ensure that the developed methods can be reliably assessed and reused by the broader research and engineering communities. As a forward-looking component, the project will also include a critical assessment of emerging quantum and hybrid quantum-classical optimization approaches within the derivative-free global optimization context, identifying their potential advantages, limitations, and future applicability for complex optimization tasks.

Funding source - The project is co-funded by the European Union under the project „My First Research Team (MPK)“ (No. 10-092-P-0001), under the joint activity agreement with the Research Council of Lithuania.