The TraDE-OPT network brought together 15 Early Stage Researchers (ESRs) across seven universities and one SME in seven different countries, creating a rich and diverse research environment at the intersection of optimization, data science, and machine learning.
To provide innovative training, we organized three schools, one workshop and one final conference, offering a comprehensive scientific and soft-skills program. The scientific sessions covered convex optimization, inverse problems, machine learning, industry presentations, an algorithmic bootcamp, a TRIZ course, equipping fellows with cutting-edge tools to tackle complex optimization problems. The soft-skills training focused on entrepreneurship, scientific writing, communication, CV writing and job market preparation, EU proposal writing, ensuring that ESRs were well-prepared for careers in both academia and industry.
The ESRs actively engaged with the global research community, presenting their findings at approximately 50 events, including conferences, workshops, and invited seminars. Their research has already resulted in around 50 publications in respected journals and international conferences, demonstrating the project’s contribution to the field. Beyond academic contributions, the TraDE-OPT consortium also prioritized public outreach, sharing its work with broader audiences to promote interest in optimization and its practical applications.
One of the major achievements of TraDE-OPT is the focus on exploiting problem structures in optimization models. Researchers developed new first-order algorithms that take advantage of structural properties to improve convergence for convex and nonconvex problems. These algorithms incorporate decomposition and parallelization techniques, making them scalable to high-dimensional data, which is essential for modern data science applications. Additionally, new randomized and incremental methods were introduced to improve computational efficiency and robustness. Machine learning techniques were integrated into optimization frameworks through unrolling techniques and hyperparameter tuning, linking data-driven models with mathematical optimization.
A key part of the project was collaboration with industry, particularly through secondments, where fellows worked directly with companies to refine their theoretical contributions and transfer optimization techniques to real-world applications. Advances in imaging science led to the development of new variational approaches for image reconstruction and segmentation, which could improve medical imaging techniques such as MRI reconstruction. Optimization methods from TraDE-OPT were successfully applied to chromatography in analytical chemistry, enhancing data processing and experimental analysis. Additionally, algorithms developed within the project contributed to the identification of railway vehicle components’ characteristics from measured data, supporting system monitoring and predictive maintenance in transportation engineering.
Through innovative research and training, international collaboration, and industry partnerships, TraDE-OPT equipped a new generation of researchers with essential skills. By integrating cutting-edge optimization techniques with interdisciplinary applications, TraDE-OPT has significantly advanced the state of the art in data-driven optimization.
