The conference took place from September 15 to 19, 2025 in Porto, Portugal, and offered great opportunities for discussing current developments and new ideas in machine learning.
Veronika highlighted the key findings of her paper Analyzing and Correcting Biased Machine Learning-Based Tuning of Weight Shrinkage in Forecast Combination in a 20-minute talk and discussed the work in more detail in the subsequent poster session.
The paper studies biases resulting from using cross-validation to tune the shrinkage level of combination weights. The study is split into three parts. First, it analyzes the impact of different data-related characteristics on tuning biases. Based on the analytical results, a bias prediction model is developed. Finally, this model is used to correct shrinkage biases, which is evaluated in the paper and shows improvements regarding the predictive accuracy in different settings.
The paper is published in Machine Learning and Knowledge Discovery in Databases, Research Track and can be found here: https://doi.org/10.1007/978-3-032-05962-8_19
Many thanks to the organizers for the inspiring event in such a wonderful city!
