Method

Tabular deep learning

Neural architectures for tabular data (TabM, TabBERT, DCN).

16 cases from 12 companies apply Tabular deep learning. 3 were checked in an online A/B test; 6 report results in production.

Typical tasks: Evaluate models, Learn user and item embeddings, Rank search results, Forecast demand and load and 6 more. Companies: Affirm, DoorDash, Criteo, Airbnb, Stripe, Zillow and 6 more. Often combined with: Gradient boosting, Multi-task learning, Logistic regression, Uncertainty quantification, Probability calibration and 41 more.