SEER: Self-Evolving Event Reasoning and Retrieval for Time Series Forecasting
SEER improves event-conditioned time-series forecasts with self-evolving retrieval and causal memory.
SEER is a closed-loop framework that conditions time-series forecasts on exogenous events rather than numerical history alone. Prediction errors drive a reflective retrieval memory that refines queries and a persistent causal knowledge base of transferable dynamics, with strict chronological boundaries to avoid look-ahead leakage. Across six volatile benchmarks, it outperforms time-series foundation models and language-model baselines.
- SEER couples event retrieval with causal reasoning for time-series forecasts.
- Errors update a retrieval memory and a persistent causal knowledge base.
- Chronological boundaries block look-ahead bias and data leakage.
- It beats time-series foundation models and language baselines on six benchmarks.
From the original article
1 min read at huggingface.co
Real-world time series are frequently driven by exogenous events and structural shifts, rendering conventional forecasting based solely on historical numerical observations insufficient. While language models can retrieve external news, standard retrieval-augmented approaches struggle with high noise, missing signals, and an inability to reason causally about event impacts.
By Mingtian Tan, Palash Goyal, Mihir Parmar · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.