BI receives research funding from UC Berkeley

FORESIG is a collaboration between BI, NTNU, and UC Berkeley. At BI, the project is managed by the research center AMOR (Center for Applied Mathematics and Operations Research), and BI professor Fabian Andsem Harang will lead it together with professor Xin Guo at UC Berkeley and professor Kurusch Ebrahimi-Fard at NTNU. Close collaboration with industry partners will be crucial for developing solutions that work in real-world settings.

“Working with Amazon in California and REMA 1000 in Norway gives us access to data and insights that allow us to test our ideas on real-world challenges. I’m glad that both companies are keen to help develop better mathematical models that can strengthen sustainability and operational efficiency in the retail sector,” says Fabian Andsem Harang, professor at BI and center director at AMOR.

The project has been awarded $25,000 through a grant from the Peder Sather Center, an international research and educational collaboration between the prestigious UC Berkeley and Norwegian universities and business schools.

“This funding means we can make a proper investment in the project and carry out essential activities such as travel, research stays and workshops – all of which are crucial to achieving academic breakthroughs and delivering solutions that industry can put into practice,” says Harang.

Developing new approaches to improve forecasting

FORESIG stands for Forecasting with Signatures: From Theory to Applications. The goal is to develop mathematical methods and forecasting models that perform better than those used today.

“Forecasts influence everything from inventory and purchasing to pricing, so accuracy really matters. Current methods often struggle when data changes quickly or is irregular, and many AI models need huge amounts of data and aren’t very user-friendly,” says Harang.

The team will use a statistical approach based on signature transformations, a mathematical framework that captures complex, non-linear patterns, copes with unstable data and performs well even when data is limited.

“This makes results easier to interpret and, in practice, leads to better demand forecasts and more dynamic pricing models that respond faster to market changes. In short, it helps reduce excess stock, avoid unnecessary waste, and prevent empty shelves,” says Harang.