Google Unveils TabFM: Pre-Trained Model for Predictive Analytics

Google has introduced a new pre-trained foundation model for tabular data called TabFM, which allows enterprise teams to generate predictions from structured data without needing to train and deploy a machine learning model from scratch.

The model, developed by Google Research, is now available in preview and can perform classification and regression tasks. Classification is used to predict categories or outcomes, while regression predicts numerical values such as customer worth or future claim amounts.

In contrast to traditional machine learning workflows, where teams would need to train and deploy separate models for each use case, TabFM uses historical data as examples to make predictions without requiring extensive training. Teams can provide labeled historical data alongside new data they want to make predictions about, using BigQuery's AI.PREDICT function.

The company is also providing an AI.EVALUATE function to assess the model's performance on prediction tasks and compare it with actual outcomes in the data.