Google Research has developed a new method to enhance time-series forecasting models, transforming their existing TimesFM model into a few-shot learner. This approach uses continued pre-training with special separator tokens to enable the model to learn from in-context examples at inference time. The resulting TimesFM-ICF model achieves performance comparable to supervised fine-tuning without the need for complex, task-specific training, making forecasting more efficient and accessible. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
RANK_REASON This is a research paper detailing a novel approach to time-series forecasting presented at ICML 2025.
