Fair Isaac Corporation

10/09/2024 | Press release | Distributed by Public on 10/09/2024 07:05

FICO’s Leadership in Optimization – Interview with Timo Berthold

FICO's team of experts in mathematical optimization drives important research that powers our leading-edge tools and capabilities, including our FICO Xpress line of products and the optimization capabilities in FICO Platform. This work advances the power of mathematical optimization, which is driving creative solutions to problems in everything from credit risk management to supply chain management to energy market modeling.

One of those experts is Timo Berthold, director of Discrete Optimization Research & Development at FICO, who was recently interviewed by Anand Subramanian's popular "Subject to" podcast series about leaders in optimization and analytics. The interview with Timo provides an in-depth look at his professional journey, experiences, and contributions to the field of mixed integer programming (MIP). Timo is recognized for his computational optimization expertise and leadership in developing advanced MIP solvers. This post gives an overview of the podcast episode - we highly recommend watching it yourself for more details and an engaging conversation:

Timo, who studied Mathematics with a minor in Computer Science at TU Berlin, developed a fascination with optimization, particularly mixed integer programming, which became a central theme of his academic and professional life. His Master's thesis, "Primal Heuristics for Mixed Integer Programs," is widely cited in the field of MIP. This thesis set the stage for his PhD research, during which he focused on algorithms for global mixed integer nonlinear programming solvers.

Timo's PhD work explored various aspects of MIP, including branching, presolving, and primal heuristics. One of his notable contributions was the introduction of the Primal Integral, a measure that became widely used in the development of MIP solvers. His dissertation also introduced the "Undercover Heuristic," a method that identifies the largest mixed integer linear subproblem within a nonlinear problem.

Bringing Technological Innovations in Optimization to FICO Xpress

After completing his PhD, Timo joined FICO, transitioning from academia to industry. He is now the team leader for Discrete Optimization at FICO Xpress. At FICO, Timo and the team he leads continue to push the boundaries of optimization technology, integrating ever-new innovative technology into Xpress.

Xpress is one of the leading commercial solvers in the optimization field, with a history spanning over four decades. Timo explains that Xpress remains competitive due to its continuous development, incorporating new algorithms, enhancing solver performance, and addressing the evolving needs of the industry. A significant achievement was the recent extension of Xpress to handle non-convex mixed integer nonlinear programming (MINLP) problems, positioning it as a pioneer among major commercial solvers capable of solving these complex problems to optimality.

Xpress uses a variety of solution methods, including outer approximation and spatial branch-and-cut algorithms, which dynamically generate linear relaxations to solve non-convex problems efficiently. Timo also addresses the importance of numerical stability in solver performance and gives advice on how modelers can effectively avoid numerical issues.

The FICO Xpress Solver and Modelling teams (Timo: standing, 2nd from left; Oliver: 4th from right)

The Role of Machine Learning and Future Directions

Machine learning plays a growing role in enhancing optimization solvers. At FICO, Timo's team leverages machine learning models to improve decision-making processes within the solver. He highlights the potential of AI to aid in specific applications but notes the inherent challenges in identifying impactful areas within MIP solvers, due to the noise and variability in performance measures.

A Leader in the Optimization World

Timo remains deeply engaged with the optimization community. He maintains close relationships with other prominent figures in MIP, reflecting the collaborative spirit within the computational optimization field. Timo also shares his personal passion for punk rock, attending over 100 concerts of his favorite German punk band, Die Ärzte.

Timo's commitment to academia continues through his role as a lecturer at TU Berlin. In this role, he recently chaired the CO@Work summer school in Berlin which featured lectures of dozens of leading optimization experts from industry and academia, and was proudly sponsored by FICO. Timo is currently working on a book about primal heuristics in MIP solvers, which is expected to be published in mid-2025. He remains enthusiastic about further advancing the field of optimization, exploring new application areas, and mentoring the next generation of researchers and developers. His work at FICO Xpress, combined with his academic pursuits, continues to shape the field of mixed integer programming, pushing the boundaries of what optimization technology can achieve.

Learn More About FICO and Computational Optimization