Reproduce “TS-Haystack: A Multi-Task Retrieval Benchmark for Long-Context Time-Series Reasoning” 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
Nicolas Zumarraga, Thomas Kaar, Ning Wang, William Tennien, Alpay Hasanli, Max Rosenblattl, Fan Wu, Kevin Riehl and 7 more
ARTS separates local signal recognition from long-recording reasoning by building a timestamped timeline, then letting an LLM inspect selected segments. TS-Haystack evaluates event-grounded questions across four time-series domains at contexts from 100 seconds to 24 hours. On two-hour Capture24 localization, ARTS reaches 0.55 accuracy versus 0.00 for ITFormer and Flamingo; this semi-synthetic setting does not isolate the agentic mechanism.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.