XBNet : An Extremely Boosted Neural Network
This addresses the challenge of enhancing neural network performance on tabular data for machine learning practitioners, though it appears incremental as it builds on existing boosted tree and neural network methods.
The paper tackles the problem of neural networks underperforming on tabular data compared to tree-based models by proposing XBNet, a novel architecture that combines tree-based models with neural networks, resulting in improved interpretability and performance through a new optimization technique called Boosted Gradient Descent for Tabular Data.
Neural networks have proved to be very robust at processing unstructured data like images, text, videos, and audio. However, it has been observed that their performance is not up to the mark in tabular data; hence tree-based models are preferred in such scenarios. A popular model for tabular data is boosted trees, a highly efficacious and extensively used machine learning method, and it also provides good interpretability compared to neural networks. In this paper, we describe a novel architecture XBNet, which tries to combine tree-based models with that of neural networks to create a robust architecture trained by using a novel optimization technique, Boosted Gradient Descent for Tabular Data which increases its interpretability and performance.