FlexTab: A Flexible Encoder-Decoder Architecture for In-Context Learning Across Diverse Tabular Tasks
FlexTab is a flexible encoder-decoder architecture designed for in-context learning on tabular data. It uses a single, task-agnostic encoder to generate general row embeddings, which are then fed into a suite of task-specific decoders. This allows the model to perform diverse tasks like classification, regression, and anomaly detection across various real-world tables without needing extensive task-specific retraining.