AI
Google releases TimeFM-3 time series forecasting model requiring no additional training
Gigazine
Key Points
- Google Research released the TimesFM-3 model as open source for time series forecasting.
- The model is pre-trained with over 1 trillion data points and requires no additional training.
- It can analyze different data types such as sales, weather, and customer counts together.
- Researchers indicate the model has broad potential for use in fields like finance, logistics, and climate.
By the Numbers
Over 1 trillion data points
Google Research announced TimesFM-3, an AI model capable of forecasting the future from multiple datasets that change over time, such as sales, customer traffic, and weather. The model has been pre-trained on a massive time series dataset containing over 1 trillion data points.
The most distinctive feature of TimesFM-3 is its "zero-shot" forecasting ability, meaning it requires no additional fine-tuning or retraining for a new task. This allows the model to process data from different domains simultaneously and produce forward-looking predictions directly.
React to this story
Ask about this story
Answers are AI-generated from this story only.
Frequently Asked Questions
- What type of data does TimesFM-3 work with?
- It works with numerical data that changes over time (sales figures, weather measurements, customer traffic, etc.).
- Is additional training required to use the model?
- No, the model does not require additional training because it is pre-trained with over 1 trillion data points to perform zero-shot forecasting.
- In which fields can this model be used?
- It can be used in many fields requiring time series analysis, such as financial forecasting, demand planning, logistics optimization, and climate modeling.
This is an AI-generated summary. The full story lives at the source.
Read the full story at the sourcegigazine.netHow we produce our content →Related stories
AI