SEER: Self-Evolving Event Reasoning and Retrieval for Time Series Forecasting
SEER uses its own forecast errors to improve what events it retrieves and how it reasons about them.
The paper frames time-series forecasting as a problem where outside events and structural shifts can matter as much as past values. SEER adds a closed loop: one memory refines future searches and filters noisy event hits, while another stores causal domain knowledge for reuse. The authors say it keeps strict chronological boundaries to avoid look-ahead bias and data leakage. Across six volatile benchmarks, they report gains over time-series foundation models and language-model baselines. HF Daily Papers' note
The paper frames time-series forecasting as a problem where outside events and structural shifts can matter as much as past values. SEER adds a closed loop: one memory refines future searches and filters noisy event hits, while another stores causal domain knowledge for reuse. The authors say it keeps strict chronological boundaries to avoid look-ahead bias and data leakage. Across six volatile benchmarks, they report gains over time-series foundation models and language-model baselines. HF Daily Papers' note
score 4