In machine learning, optimization is part of the fundamental toolbox for enabling the large language models we are using today. The ELLIIT focus period on optimization for learning will gather visiting scholars in Lund from mid-April to mid-May, to address optimization problems, and to broaden collaboration.
The presence of large language models in everyday digital life has become normalized. But what makes the models so increasingly good at predicting the next string of letters in its answers to our daily queries?
”One part of the training of large language models is optimization. Before the models are deployed, and released to use, they are trained on immense amounts of data. The optimization algorithms are tools to mathematically reach the best solution for your model out of all the choices”, says Pontus Giselsson, associate professor at Lund University, and ELLIIT focus period organizer.
Optimization is a widely used mathematical tool with application areas that extend well beyond machine learning and large language models. It plays an important role in areas such as automatic control, computer vision, signal processing,
and statistics.
”Optimization algorithms adjust the parameters of the model in question, to minimize errors, and in effect result in a model with the best predictions”, Pontus Giselsson explains.
Proving your algorithm
The focus period in Lund will gather visiting scholars from different areas of the optimization field, with practical and theoretical approaches to learning. A central question is analysis of optimization algorithms and guarantees of convergence to an optimal solution, and in effect a well performing model in the machine learning application.
“There is an interplay of practical approaches and optimization analysis, and for the focus period symposium we gather experts to share insights on theory, algorithms, and practice”, says Pontus Giselsson.
For the visiting scholars, Pontus Giselsson together with organizers hope to create an open and allowing environment, for discussions and new collaborations to form:
“We encourage interactions between visiting scholars and our research environment here in Lund, and hope to enable researchers to really take the time to work on these problems together”, says Pontus Giselsson.
The focus period on Optimization for Learning takes place in Lund April 20 – May 22, with the three-day symposium May 6-8, 2026.
Optimization problems are sometimes visualized as mountain ranges, where the solution is the lowest point in one of the valleys. To find it, the algorithm only has access to local slope information, like hiking in fog: you can’t see the whole route, you just feel which way is downhill and take the next step. And unlike a real mountain range, the optimization landscape is high-dimensional, sometimes spanning millions to billions of parameters. Image: Pixabay/Erik Tanghe.
Focus Period Optimization for Learning
Read more about the focus period on Optimization for Learning on the focus period main page.
Register for Symposium
See the confirmed invited speakers for the three day symposium May 6 – 8, 2026
and register for the symposium in Lund.
Visiting Scholars
Meet the visiting scholars who will attend the focus period and symposium in Lund.