Training-Free Continual Learning for Multimodal Forecasting via Prompt Evolution

Key points

• Adapting a forecasting system to a changing world classically means retraining.1
• On a refreshed multimodal benchmark (weekly search indices, daily currencies, daily commodities, each window paired with web-retrieved news), we run GEPA-style prompt evolution under strict contamination control to prevent the data leakage common in timeseries benchmarks: the forecaster's knowledge cutoff predates every evaluated window.1
• A 900-word evolved guidance carried by Llama-3.3-70B beats TimeSeries Foundation Models (TSFMs) on held-out data (GM MASE 0.776 vs. 0.898 for TabPFN-TS-3), transfers to more recent models without measurable loss, and costs dollars rather than GPU-days to update.1
• We argue that continual prompt evolution is a practical, training-free alternative to retraining for multimodal forecasting.1

Sources (1)

• [1]Training-Free Continual Learning for Multimodal Forecasting via Prompt Evolution
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 01:38 AM

  We ask whether the task of continually learning regimes for time series forecasting can be represented by a piece of text: a guidance block, evolved by outcome-driven reflection and used in a frozen LLM forecaster's prompt.

  Adapting a forecasting system to a changing world classically means retraining.

Extractive summary: sentences quoted from the sources.