Summary of Graph-based Forecasting with Missing Data Through Spatiotemporal Downsampling, by Ivan Marisca et al.
Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling
by Ivan Marisca, Cesare Alippi, Filippo Maria Bianchi
First submitted to arxiv on: 16 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Spatiotemporal forecasting involves predicting future observations for each point in a set of synchronous time series, taking into account inter-series relationships. Spatiotemporal graph neural networks have achieved impressive results by representing these relationships as graphs, but existing methods often rely on unrealistic assumptions about input availability and fail to capture hidden dynamics when data is missing. This work tackles this problem through hierarchical spatiotemporal downsampling, which progressively coarsens the input time series over time and space. The resulting representations are combined using an attention mechanism conditioned on observations and missing data patterns to generate forecasts. Our approach outperforms state-of-the-art methods on synthetic and real-world benchmarks under different missing data distributions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Spatiotemporal forecasting is like predicting what will happen at different points in space and time, based on how things are related across time series. The current best methods do a great job of this, but they have some limitations. What if not all the information is available? That’s where our new approach comes in. We use something called hierarchical spatiotemporal downsampling to reduce the amount of data we need and focus on what’s really important. Then, we use an attention mechanism to combine this reduced data with the information we do have to make predictions. Our method works better than others on real-world and made-up data when some of the data is missing. |
Keywords
* Artificial intelligence * Attention * Spatiotemporal * Time series
