Descent-Net: Learning Descent Directions for Constrained Optimization

2025/12/12 by Zhou, Zisheng, Zheng, Dengyu, Chen, Zirui +1
Computer Science · Engineering · #Advanced Neural Network Applications #Electric Power System Optimization #FOS: Mathematics #Optimal Power Flow Distribution #Optimization and Control (math.OC)

paper · doi:10.48550/arxiv.2512.11396

openalex publication_date 2025/12/12 · openalex created_date 2025/12/16 · openalex updated_date 2026/07/28

Abstract

Deep learning approaches, known for their ability to model complex relationships and fast execution, are increasingly being applied to solve large optimization problems. However, existing methods often face challenges in simultaneously ensuring feasibility and achieving an optimal objective value. To address this issue, we propose Descent-Net, a neural network designed to learn an effective descent direction from a feasible solution. By updating the solution along this learned direction, Descent-Net improves the objective value while preserving feasibility. Our method demonstrates strong performance on both synthetic optimization tasks and the real-world AC optimal power flow problem, while also exhibiting effective scalability to large problems, as shown by portfolio optimization experiments with thousands of assets.

Citations

• HPR-QP: A dual Halpern Peaceman-Rachford method for solving large-scale convex composite quadratic programming
• PDHG-Unrolled Learning-to-Optimize Method for Large-Scale Linear Programming
• Clarabel: An interior-point solver for conic programs with quadratic objectives
• Learning to Optimize: A Primer and A Benchmark
• Learning Linear Programs from Optimal Decisions
• Projection-Based Constrained Policy Optimization
• Differentiable Convex Optimization Layers
• Deep Equilibrium Models
• Why gradient clipping accelerates training: A theoretical justification for adaptivity
• Learning Warm-Start Points for AC Optimal Power Flow
• Differentiable Linearized ADMM
• Machine Learning for Combinatorial Optimization: a Methodological Tour d'Horizon
• Machine learning for combinatorial optimization: A methodological tour d’horizon
• Theoretical Linear Convergence of Unfolded ISTA and its Practical Weights and Thresholds
• Neural Ordinary Differential Equations
• OptNet: Differentiable Optimization as a Layer in Neural Networks
• Learning to learn by gradient descent by gradient descent
• Generating Sequences With Recurrent Neural Networks
• On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming

Related

• Enhancing Deep Learning with Optimized Gradient Descent: Bridging Numerical Methods and Neural Network Training
• Learned Optimizers that Scale and Generalize
• FSNet: Feasibility-Seeking Neural Network for Constrained Optimization with Guarantees
• Make Optimization Once and for All with Fine-grained Guidance
• Learning to Optimize Neural Nets