Tag
A new moving-horizon approximate branch-and-reduce method for training near-optimal deep classification trees on large-scale datasets with continuous features, achieving better accuracy than heuristic baselines and far greater scalability than global optimal solvers.