Frontiers of Foundation Models for Time Series
summary
Claims a practical path to foundation models for time series by building Temple — a prompt-based generative pretraining Transformer that decomposes inputs into trend/seasonality/residuals, feeds soft prompts into a Transformer backbone, and delivers zero-shot forecasting that matches or beats state-of-the-art models trained on target domains (benchmarks: electricity, weather, New York taxi, finance, social news). Adds a diffusion + Transformer generative model for multivariate imputation and uncertainty estimation that outperforms baselines on MIMIC-III and synthetic tests; code lands on GitHub and Hugging Face.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 0:15The AI Era: A Pivotal Moment
- 5:34Fundamental AI - Pushing the Frontier of AI
- 9:46Practical Challenges: Real-world Time Series Data
- 19:59Foundation Models for Time Series | Contextual Information
- 27:11Foundation Models for Spatiotemporal Data
related talks
31mFoundational Models for Time Series Forecasting: are we there yet?
31mFoundational Time Series Models in Practice: The Future of Forecasting, or Just Hype?
21mFoundational Models for Time Series Forecasting: are we there yet?
19mForecasting for Everyone - Foundation Models and the Future of Time Series Prediction
30mFoundational Models for Time Series Forecasting: are we there yet
31mAdvanced machine learning time series....
54mTransformes for Time Series: Is the New State of the Art (SOA) Approaching?
10mWhy Your Forecasting Transformer Isn’t Working (And How To Fix It...