TabICLv2
A large-scale Prior-Fitted Network (PFN)-style model for tabular data, designed for tabular prediction tasks under the In-Context Learning paradigm.
Timeline (last 90 days)
The Structured Data Models library currently wraps three types of models: TabICLv2 and KumoTabular for single tables, and KumoRelational for relational databases.
TabICLv2, KumoTabular, and KumoRelational all use in-context learning for prediction, taking training samples as context at inference time without gradient updates.
TabICLv2 is an iterated version of the TabICL (Tabular In-Context Learning) series of models, whose core is pretraining a Transformer on a large number of synthetic tabular tasks so that it completes classification or regression relying on in-context examples.
The improvement from Bayesian Model Averaging also holds for the largest-scale TabICLv2 model.
TabICLv2 is currently one of the larger-scale tabular in-context learning foundation models.
All Facts (5)
The Structured Data Models library currently wraps three types of models: TabICLv2 and KumoTabular for single tables, and KumoRelational for relational databases.
50%UnverifiedTabICLv2, KumoTabular, and KumoRelational all use in-context learning for prediction, taking training samples as context at inference time without gradient updates.
50%UnverifiedTabICLv2 is an iterated version of the TabICL (Tabular In-Context Learning) series of models, whose core is pretraining a Transformer on a large number of synthetic tabular tasks so that it completes classification or regression relying on in-context examples.
50%UnverifiedThe improvement from Bayesian Model Averaging also holds for the largest-scale TabICLv2 model.
50%UnverifiedTabICLv2 is currently one of the larger-scale tabular in-context learning foundation models.
50%