Large pre-trained neural networks specifically engineered to process, analyze, and forecast sequential time-stamped data points, such as financial ticker prices, sensor logs, or traffic patterns.

Brandished by fintech and logistics vendors to convince clients that standard linear regression and statistical forecasting tools have been replaced by a sentient time-predicting supercomputer.

• Futuristic Magic Claims: Promising an off-the-shelf model can accurately predict complex real-world market trends simply because it was pre-trained on generic numbers.
• Over-Engineered Vocabulary: Rebranding basic statistical trend analysis as a "foundation model."
• House-Cleaning Reality Check: Like tracking what time you take out the trash each week and calling your calendar a "time series foundation model."

Margin Call (2010s) scene where the young analyst pulls up historical volatility models on his computer screen, watching time-series data curves signal imminent financial disaster.

A pre-trained software model designed to spot patterns and trends in time-stamped data.

“Time series foundation models allow our system to anticipate inventory shortages three weeks early.”

“We used a pre-trained time-series model to analyze our historical website traffic logs and forecast server capacity needs.”