Eilam Shapira, Alan Arazi, Shoham Grunblat, Roi Reichart (Technion) · Target-aware tabular encoding

LightTAR: Light and Fast Target-Aware Representations for Multimodal Tabular Learning

LightTAR augments standard PCA reduction of pretrained text and image encoder embeddings with a few target-aware features, extracted from the final and an intermediate encoder layer using regularised partial least squares. The authors evaluate it with 21 tabular models, including tree ensembles, neural networks and tabular foundation models, and report improved average predictive performance across all of them. The gains come with a small increase in runtime over PCA, and the method needs no changes to the encoder or the learner.

Open the source ↗

Similar Research & Safety releases

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.