Quantum-Train-Based Distributed Multi-Agent Reinforcement Learning

  • 2024-12-12 00:51:41
  • Kuan-Cheng Chen, Samuel Yen-Chi Chen, Chen-Yu Liu, Kin K. Leung
  • 0

Abstract

In this paper, we introduce Quantum-Train-Based Distributed Multi-AgentReinforcement Learning (Dist-QTRL), a novel approach to addressing thescalability challenges of traditional Reinforcement Learning (RL) byintegrating quantum computing principles. Quantum-Train Reinforcement Learning(QTRL) leverages parameterized quantum circuits to efficiently generate neuralnetwork parameters, achieving a \(poly(\log(N))\) reduction in thedimensionality of trainable parameters while harnessing quantum entanglementfor superior data representation. The framework is designed for distributedmulti-agent environments, where multiple agents, modeled as Quantum ProcessingUnits (QPUs), operate in parallel, enabling faster convergence and enhancedscalability. Additionally, the Dist-QTRL framework can be extended tohigh-performance computing (HPC) environments by utilizing distributed quantumtraining for parameter reduction in classical neural networks, followed byinference using classical CPUs or GPUs. This hybrid quantum-HPC approach allowsfor further optimization in real-world applications. In this paper, we providea mathematical formulation of the Dist-QTRL framework and explore itsconvergence properties, supported by empirical results demonstratingperformance improvements over centric QTRL models. The results highlight thepotential of quantum-enhanced RL in tackling complex, high-dimensional tasks,particularly in distributed computing settings, where our framework achievessignificant speedups through parallelization without compromising modelaccuracy. This work paves the way for scalable, quantum-enhanced RL systems inpractical applications, leveraging both quantum and classical computationalresources.

 

Quick Read (beta)

loading the full paper ...