Reproduce “CoSTI: Consistency Models for (a faster) Spatio-Temporal Imputation” 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
Javier Solís-García, Belén Vega-Márquez, Juan A. Nepomuceno, Isabel A. Nepomuceno-Chamorro
Researchers at the University of Seville introduced CoSTI, the first adaptation of Consistency Models for Multivariate Time Series Imputation, which achieves imputation accuracy comparable to diffusion-based methods while reducing inference time by up to 98%. This enables efficient data recovery for real-time spatio-temporal applications.
Key insights
- Consistency Models can be successfully adapted from computer vision to the Multivariate Time Series Imputation domain, enabling fast generative modeling for time series data.
- Incorporating comprehensive conditional information, including interpolated observations, graph structure, and missing masks, is critical for effective spatio-temporal data imputation.
- A curriculum learning strategy that progressively increases the number of noise levels during training enhances model robustness and imputation accuracy for consistency models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.