Foundational Models for Time Series Forecasting: are we there yet
summary
Claims center on Amazon Chronos and Salesforece's Moirai as transformer-based, zero-shot forecasters trained on massive pooled corpora (billions to tens of billions of observations) using methods like T5 re‑training with quantized tokens, patch tokenization, and 5–10% synthetic data augmentation. Papers report scaling benefits only when hundreds of thousands of series exist, show easy probabilistic outputs via token logits, and warn benchmarks can be contaminated so practical production gains remain conditional.
AI-written summary · about
chapters (9)
- 0:00Time series forecasting models
- 2:26Transformers in production
- 9:04Data scarcity limits
- 13:39Foundational model architectures
- 17:30Chronos training
- 19:36Times FM approach
- 24:55Benchmarking challenges
- 26:20Adoption recommendations
- 27:02Q&A
related talks
21mFoundational Models for Time Series Forecasting: are we there yet?
31mFoundational Models for Time Series Forecasting: are we there yet?
31mFoundational Time Series Models in Practice: The Future of Forecasting, or Just Hype?
19mForecasting for Everyone - Foundation Models and the Future of Time Series Prediction
33mFrontiers of Foundation Models for Time Series
40mMulti-Series Forecasting at Scale with StatsForecast
31mAdvanced machine learning time series....
29mProbabilistic demand forecasting at scale