Abstract:For a general purpose optimization problem over a finite rectangle region, this paper pioneers a unified slot machine framework for global optimization by transforming the search for global optimizer(s) to the optimal strategy formulation of a bandit process in infinite policy sets and proves that two-armed bandit is enough. By leveraging the strategic bandit process-driven optimization framework, we introduce a new {\bf S}trategic {\bf M}onte {\bf C}arlo {\bf O}ptimization (SMCO) algorithm that coordinate-wisely generates points from multiple paired distributions and can be implemented parallel for high-dimensional continuous functions. Our SMCO algorithm, equipped with tree search that broadens the optimal policy search space of slot machine for attaining the global optimizer(s) of a multi-modal function, facilitates fast learning via trial and error. We provide a strategic law of large numbers for nonlinear expectations in bandit settings, and establish that our SMCO algorithm converges to global optimizer(s) almost surely. Unlike the standard gradient descent ascent (GDA) that uses a one-leg walk to climb the mountain and is sensitive to starting points and step sizes, our SMCO algorithm takes a two-leg walk to the peak by using the two-sided sampling from the paired distributions and is not sensitive to initial point selection or step size constraints. Numerical studies demonstrate that the new SMCO algorithm outperforms GDA, particle swarm optimization and simulated annealing in both convergence accuracy and speed. Our SMCO algorithm should be extremely useful for finding optimal tuning parameters in many large scale complex optimization problems.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to develop a computationally fast global optimization algorithm to deal with the problem of searching for the global optimal solution of high - dimensional continuous functions (which may have multiple extreme points). Such problems are very common in practical applications, but traditional optimization methods such as Gradient Descent Ascent (GDA), Particle Swarm Optimization (PSO) and Simulated Annealing have problems such as slow convergence speed and being easily trapped in local optima when dealing with high - dimensional, multi - modal, non - convex functions. ### Core contributions of the paper 1. **Unified two - armed bandit framework**: This paper proposes a unified "two - armed bandit" framework, which transforms the global optimization problem into a two - armed bandit problem with an infinite strategy set. Through this framework, the author proves that a two - armed bandit is sufficient to achieve effective global optimization. 2. **Strategic Monte Carlo Optimization (SMCO) algorithm**: Based on the above framework, the author introduces a new optimization algorithm - Strategic Monte Carlo Optimization (SMCO). This algorithm generates samples coordinate - by - coordinate from multiple paired distributions and can process high - dimensional continuous functions in parallel. Compared with traditional methods, the SMCO algorithm does not depend on the selection of initial points or step - length constraints, thus improving robustness and efficiency. 3. **Theoretical guarantees**: The paper provides the Law of Large Numbers in the two - armed bandit setting for non - linear expectations and proves that the SMCO algorithm converges almost surely to the global optimal solution. In addition, the convergence rate of the SMCO algorithm is established, further verifying its effectiveness and superiority. 4. **Tree - search extension**: In order to further improve the global optimization ability, the author introduces Monte Carlo Tree Search (MCTS) into the SMCO algorithm to form SMCO with Tree Search. This method effectively balances exploration and exploitation by iteratively expanding the search tree, ensuring that the algorithm can escape from local optima and find the global optimal solution. ### Specific application scenarios This algorithm has broad application prospects in many large - scale and complex optimization problems, such as finding optimal tuning parameters. Through numerical experiments, the paper shows that the SMCO algorithm is superior to existing methods such as GDA, PSO and Simulated Annealing in terms of convergence accuracy and speed. In conclusion, this paper proposes a brand - new optimization framework and algorithm by transforming the global optimization problem into a two - armed bandit problem, providing a powerful tool for solving the global optimization problems of high - dimensional, multi - modal, non - convex functions.

Solving a global optimal problem requires only two-armed slot machine

Several Algorithms of Global Optimal Search

A new hybrid genetic algorithm for global minimax optimization

SMGO-$Δ$: Balancing Caution and Reward in Global Optimization with Black-Box Constraints

Globally Optimized Matchmaking in Online Games

A novel batch-selection strategy for parallel global optimization

Monte Carlo Policy Gradient Method for Binary Optimization

Genetic multi-armed bandits: a reinforcement learning approach for discrete optimization via simulation

Towards One-shot Neural Combinatorial Solvers: Theoretical and Empirical Notes on the Cardinality-Constrained Case

Adaptive nested Monte Carlo approach for multi-objective efficient global optimization

Non-stationary Continuum-armed Bandits for Online Hyperparameter Optimization.

Surrogate-guided multi-objective optimization (SGMOO) using an efficient online sampling strategy

Generalizable Meta-Heuristic based on Temporal Estimation of Rewards for Large Scale Blackbox Optimization

Global Optimization: A Machine Learning Approach

A Selective Biogeography-Based Optimizer Considering Resource Allocation for Large-Scale Global Optimization

A multi-strategy enhanced sine cosine algorithm for global optimization and constrained practical engineering problems

A Comprehensive Multi-Strategy Enhanced Biogeography-Based Optimization Algorithm for High-Dimensional Optimization and Engineering Design Problems

A Multi-Objective Optimization Problem Solving Method Based on Improved Golden Jackal Optimization Algorithm and Its Application

Parallelized multiobjective efficient global optimization algorithm and its applications

A Competitive and Cooperative Swarm Optimizer for Constrained Multiobjective Optimization Problems

An Asymptotically Optimal Multi-Armed Bandit Algorithm and Hyperparameter Optimization