Outstanding Paper Award at ICML 2025

The Value of Prediction in Identifying the Worst-Off

Unai Fischer Abaigar, Prof. Dr. Christoph Kern, and Juan Carlos Perdomo (Harvard, NYU) received one of only six Outstanding Paper Awards at the 42nd International Conference on Machine Learning (ICML), held in Vancouver in July 2025.

Their paper, The Value of Prediction in Identifying the Worst-Off, addresses a fundamental question in the use of machine learning for public resource allocation: When is it actually worth investing in better prediction? Combining theoretical analysis with administrative data on long-term unemployment in Germany, the authors compare the benefits of improving predictive accuracy to those of expanding bureaucratic capacity.

Check out the paper here (https://arxiv.org/abs/2501.19334) and the oral presentation here (https://icml.cc/virtual/2025/session/46923)