Structured Foundational Models
Do time-series foundation models learn the structure of the world—or only fit its curves?
A large share of the world's data is time-series and tabular — stock prices, sensor readings, supply chains, weather, energy consumption, patient vitals — an opportunity estimated by Forbes at $600 billion. Our researchers are probing whether time-series foundation models genuinely learn the structure of the world or simply fit curves — work that led to our first accepted paper at the NeurIPS 2025 Workshop on Recent Advances in Time Series Foundation Models (BERT2S).


