Neural Regression Trees
This addresses a specific bottleneck in regression methods for machine learning practitioners, though it appears incremental as it builds on existing RvC approaches.
The paper tackles the suboptimal discretization in Regression-via-Classification by proposing a neural regression tree model that jointly learns optimal thresholds and features, achieving state-of-the-art results on two challenging regression tasks.
Regression-via-Classification (RvC) is the process of converting a regression problem to a classification one. Current approaches for RvC use ad-hoc discretization strategies and are suboptimal. We propose a neural regression tree model for RvC. In this model, we employ a joint optimization framework where we learn optimal discretization thresholds while simultaneously optimizing the features for each node in the tree. We empirically show the validity of our model by testing it on two challenging regression tasks where we establish the state of the art.