Machine Learning Algorithms & Natural Language Processing

Feb 9, 2026 · Artificial Intelligence

Time‑o1: Overcoming Time‑Series Forecasting Bottlenecks with a Novel Loss Function

The paper identifies two fundamental issues in time‑series forecasting—label autocorrelation bias and task‑scale explosion caused by the standard TMSE loss—and proposes Time‑o1, a PCA‑based orthogonal label transformation that eliminates bias, reduces optimization complexity, and yields consistent performance gains across multiple models and datasets.