Simons: Learning Across Time
Simons explores shared representations for time-series and tabular data, with a focus on learning how systems evolve across different settings.
Learning across temporal systems
Demand, weather, energy and operational data describe different systems, but they share a central question: how does the present become the future?
Simons explores representations that bring time-series and tabular observations into a shared modeling framework.
More than the next value
A useful temporal representation should capture relationships across variables, changes of regime and uncertainty about future outcomes. Forecasting is one way to examine those representations.
Testing across datasets
We would assess these representations on held-out datasets, different sampling frequencies and changes of regime. Forecasting quality, uncertainty and transfer would be examined separately, with the conditions of each test made explicit.