Foundational Models for Time Series Forecasting: are we there yet?
summary
Explains that transformer-based foundational forecasters are promising but data-limited: teams adapt T5-style and decoder transformers, bin or patch continuous series, and reuse Hugging Face tooling; Google/industry-scale efforts claim trained on ~100B observations but public datasets are far smaller. Papers show 10–15% synthetic data boosts performance and that patch size ~32 balances accuracy and length. Benchmarks are often contaminated by training overlap, so results look good on paper but real zero-shot reliability is still unproven.
AI-written summary · about
chapters (6)
- 0:00Introduction and motivation
- 1:09Transformers in time series forecasting
- 6:42Data scarcity in foundational models
- 13:30Model architectures overview
- 24:28Evaluation and real-world use
- 25:41Q&A session
related talks
31mFoundational Time Series Models in Practice: The Future of Forecasting, or Just Hype?
21mFoundational Models for Time Series Forecasting: are we there yet?
30mFoundational Models for Time Series Forecasting: are we there yet
33mFrontiers of Foundation Models for Time Series
19mForecasting for Everyone - Foundation Models and the Future of Time Series Prediction
31mAdvanced machine learning time series....
10mWhy Your Forecasting Transformer Isn’t Working (And How To Fix It...
54mTransformes for Time Series: Is the New State of the Art (SOA) Approaching?