Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Machine Learning Algorithms for Improving Exact Classical Solvers in Mixed Integer Continuous Optimization

Morteza Kimiaei, Vyacheslav Kungurtsev, Brian Olimba

TL;DR

The paper addresses the hardness of INLP/MINLP/CNLP by surveying how ML and RL can augment exact branch-and-bound solvers without compromising global optimality. It introduces a unified BB framework that embeds learning into branching, cut selection, node ordering, and parameter control, and surveys neural and reinforcement-based approaches across solver components. Key contributions include a solver-centric taxonomy, synthesis of prior ML/RL work, and mappings to high-impact applications along with open challenges in generalization and scalability. The work demonstrates how learning-augmented BB can accelerate convergence while preserving correctness, offering a pathway toward production-ready intelligent solvers for large-scale MINLP and CNLP problems.

Abstract

Integer and mixed-integer nonlinear programming (INLP, MINLP) are central to logistics, energy, and scheduling, but remain computationally challenging. This survey examines how machine learning and reinforcement learning can enhance exact optimization methods-particularly branch-and-bound (BB)-without compromising global optimality. We cover discrete, continuous, and mixed-integer formulations, and highlight applications such as vehicle routing, hydropower planning, and crew scheduling. We introduce a unified BB framework that embeds learning-based strategies into branching, cut selection, node ordering, and parameter control. Classical algorithms are augmented using supervised, imitation, and reinforcement learning models to accelerate convergence while maintaining correctness. We conclude with a taxonomy of learning methods by solver class and learning paradigm, and outline open challenges in generalization, hybridization, and scaling intelligent solvers.

Machine Learning Algorithms for Improving Exact Classical Solvers in Mixed Integer Continuous Optimization

TL;DR

The paper addresses the hardness of INLP/MINLP/CNLP by surveying how ML and RL can augment exact branch-and-bound solvers without compromising global optimality. It introduces a unified BB framework that embeds learning into branching, cut selection, node ordering, and parameter control, and surveys neural and reinforcement-based approaches across solver components. Key contributions include a solver-centric taxonomy, synthesis of prior ML/RL work, and mappings to high-impact applications along with open challenges in generalization and scalability. The work demonstrates how learning-augmented BB can accelerate convergence while preserving correctness, offering a pathway toward production-ready intelligent solvers for large-scale MINLP and CNLP problems.

Abstract

Integer and mixed-integer nonlinear programming (INLP, MINLP) are central to logistics, energy, and scheduling, but remain computationally challenging. This survey examines how machine learning and reinforcement learning can enhance exact optimization methods-particularly branch-and-bound (BB)-without compromising global optimality. We cover discrete, continuous, and mixed-integer formulations, and highlight applications such as vehicle routing, hydropower planning, and crew scheduling. We introduce a unified BB framework that embeds learning-based strategies into branching, cut selection, node ordering, and parameter control. Classical algorithms are augmented using supervised, imitation, and reinforcement learning models to accelerate convergence while maintaining correctness. We conclude with a taxonomy of learning methods by solver class and learning paradigm, and outline open challenges in generalization, hybridization, and scaling intelligent solvers.

Paper Structure

This paper contains 89 sections, 2 theorems, 77 equations, 10 figures, 5 tables, 14 algorithms.

Table of Contents

  1. Introduction
  2. Background and Motivation
  3. Problem Setting and Research Question
  4. Challenges in Exact (M)INLP Solving
  5. Opportunities for Machine Learning Integration
  6. Scope and Methodology
  7. Contributions of This Survey
  8. Background - Integer and Mixed Integer Programming Problems
  9. Foundation — Integer Structures
  10. Integer Nonlinear Optimization (INLP) problem
  11. Mixed-Integer Nonlinear Programming (MINLP) problem
  12. Exact Classical Methods for Integer, Continuous, and Mixed-Integer Optimization
  13. Preliminaries and Terminology
  14. Node Selection Strategies
  15. Feasible Region
  16. ...and 74 more sections

Key Result

Lemma B.1

Let $\{L_k\}$ and $\{U_k\}$ be the lower and upper bounds computed by a BB algorithm for the MINLP problem e.MINLP, and define Then the quantity is called the global optimality gap. It satisfies where $x^*$ is the best feasible solution found (attaining $U^*$) and $f^* = \inf_{x \in C_{\mathop{\rm mi}}} f(x)$ is the global minimum.

Figures (10)

  • Figure 1: Flowchart Summary of Survey Papers on ML and RL Methods for ILP, MILP, MINLP, and CNLP Problems.
  • Figure 2: Best-Bound Search (BBS): Selects the node with the lowest lower bound from the active pool. This is a best-first strategy.
  • Figure 3: Depth-First Search (DFS): Selects the most recently added node using a Last-In, First-Out (LIFO) stack. This explores one path deeply before backtracking.
  • Figure 4: Breadth-First Search (BFS): Selects the oldest node using a First-In, First-Out (FIFO) queue. This explores all nodes at the current depth before going deeper.
  • Figure 5: Evolution of the Feasible Region $\mathcal{S}_k$: Starting from continuous relaxation, each region is selected, updated, and possibly split.
  • ...and 5 more figures

Theorems & Definitions (6)

  • Definition B.1: $\epsilon$-Global Optimal Solution
  • Remark B.1
  • Lemma B.1: Global Optimality Gap
  • proof
  • Theorem B.1: Global Convergence of Enhanced BB Algorithms for MINLP
  • proof