Researchers have proposed a new evaluation metric for irregular time-series forecasting called Continuous-time Squared Error (CSE). They argue that the traditional mean squared error (MSE) metric is biased because it's influenced by the sampling distribution of timestamps. The authors claim that CSE can more accurately assess models' continuous-time predictive performance, and they provide a benchmark dataset to test this. Their experiments show that using CSE instead of MSE leads to better results in real-world scenarios. --- Why it matters: This matters for AI engineers because it highlights the importance of choosing the right evaluation metric for time-series forecasting tasks. Using an inaccurate metric can lead to biased model performance assessments, which can have significant consequences in applications such as finance and healthcare. Source: https://arxiv.org/abs/2608.17293