TW3Cast：冻结路由的轻量微调基础模型用于时间序列预测
原标题：TW3Cast: A Frozen Router of Lightly Fine-Tuned Foundation Models for Time-Series Forecasting on GIFT-Eval, Selected Entirely on the Training Split
AI 摘要
该论文提出 TW3Cast，一个用于时间序列预测的路由系统，在 GIFT-Eval 基准上以平均 MASE 排名位列 130 个条目中的第 3 名。它不使用任何智能体或语言模型，而是通过一个在训练集上一次性计算并冻结的路由表，为 97 个配置选择专家（轻量微调的 Chronos-2、TiRex、Toto 等基础模型）或混合策略。系统采用双重准确率与校准标准、非对称防记忆边际等保护机制，最佳基础模型单独使用排名 33.8，完整路由器达到 19.4。
正文节选
TW3Cast: A Frozen Router of Lightly Fine-Tuned Foundation Models for Time-Series Forecasting on GIFT-Eval, Selected Entirely on the Training SplitThanks: Every leaderboard table, figure and number of this paper regenerates by script from the released routing table, the released expert index and a dated snapshot of the public per-configuration scores; see Section 12. Abstract TW3Cast is a time-series forecasting system that reaches position 3 of 130 entries on the GIFT-Eval benchmark by mean MASE