Reproduce “$\textbf{S}^2$IP-LLM: Semantic Space Informed Prompt Learning with LLM for Time Series Forecasting” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zijie Pan, Yushan Jiang, Sahil Garg, Anderson Schneider, Yuriy Nevmyvaka, Dongjin Song
S²IP-LLM forecasts time series by decomposing inputs into trend, seasonal, and residual components, patching them separately, and retrieving similar learned anchors from vectors derived from GPT-2 word-token embeddings. It prepends the anchors as soft prompts to GPT-2. On M4 horizons 6 to 48, its weighted-average OWA is 0.857, versus 0.869 for PatchTST; this shows competitive benchmark performance, not broad generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.