Decision tree, despite its unmatched interpretability and lightweight structure, faces two key issues that limit its broader applicability: non-differentiability and low testing accuracy. This study addresses these issues by developing a differentiable oblique tree that optimizes the entire tree using gradient-based optimization. We propose an exact reformulation of hard-split trees based on "ReLU+Argmin" mechanism, and then cast the reformulated tree training as an unconstrained optimization task. The ReLU-based sample branching, expressed as exact-zero or non-zero values, preserve a unique decision path, in contrast to soft decision trees with probabilistic routing. The subsequent Argmin operation identifies the unique zero-violation path, enabling deterministic predictions. For effective gradient flow, we approximate Argmin behaviors by scaling softmin function. To ameliorate numerical instability, we propose a warm-start annealing scheme that solves multiple optimization tasks with increasingly accurate approximations. This reformulation alongside distributed GPU parallelism offers strong scalability, supporting 12-depth tree even on million-scale datasets where most baselines fail. Extensive experiments demonstrate that our optimized tree achieves a superior testing accuracy against 14 baselines, including an average improvement of 7.54\% over CART.

In this article, two algorithms are proposed to solve multiobjective path-constrained dynamic optimization problems. In each algorithm, an adaptive$\varepsilon $-constraint method is employed to solve the multiobjective dynamic optimization problems (MODOPs) with path constraints in two iterative loops. In the outer loop, the adaptive$\varepsilon $-constraint method adaptively adjusts the choice of the parameters$\varepsilon $, which transfers MODOP into a sequence of single-objective dynamic optimization problems (SODOPs) with extra inequality constraints. In the inner loop, two different algorithms are used to solve the single-objective optimization problems. The first algorithm guarantees that the path constraints can be satisfied with any finite prescribed tolerance by replacing path constraints with a finite number of point constraints. Furthermore, the second algorithm guarantees that the path constraints are rigorously satisfied by enforcing the path constraints at a limited number of time points and by restricting the right-hand side of the path constraints. The proposed algorithms are proven to converge within finite iterations. The effectiveness of the algorithms is verified via numerical studies, along with a comparison to a state-of-the-art algorithm.