Reproduce “Querying Labeled Time Series Data with Scenario Programs” 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
Edward Kim, Devan Shanker, Varun Bharadwaj, Hongbeen Park, Jinkyu Kim, Hazem Torfah, Daniel J Fremont, Sanjit A Seshia
A method developed by researchers at UC Berkeley and collaborators enables querying labeled time series data using formal scenario programs, bridging the sim-to-real gap for autonomous systems. The approach converts SCENIC programs into Hierarchical Finite State Machines, achieving an average accuracy of 0.80 and query times of 0.06 seconds, surpassing leading Vision Language Models in performance for complex scenario validation.
Key insights
- Formal methods, specifically the translation of scenario programs to HFSMs, enable highly accurate and efficient querying of complex temporal behaviors in labeled time series data.
- The approach demonstrates superior accuracy (0.80 vs. 0.45-0.60) and significantly faster execution (0.06s vs. 6-41s) compared to state-of-the-art Vision Language Models for scenario matching.
- The querying algorithm exhibits linear scalability with respect to video duration but experiences exponential growth in runtime with an increasing number of objects in the SCENIC program, primarily due to the combinatorial object correspondence search.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.