Harnessing Parallel Chaos from a Multimode Semiconductor Laser with Intensity-Modulated Optical Injection for Decision Making

The Multi-Armed Bandit (MAB) problem, central to reinforcement learning-based decision-making, involves maximizing rewards across multiple uncertain choices. As the number of decision arms increases, traditional computational methods struggle with the dual challenge of balancing exploration and exploitation while maintaining effective decision-making. In this paper, we present an innovative all-optical decision-making framework, utilizing intensity-modulated optical injection into a single multimode laser to generate multiple parallel chaotic signals. This method overcomes the limitations of conventional chaos generation, enabling parallelism with minimal correlation and without introducing delay characteristics, a common challenge in traditional chaotic systems. These parallel chaos with wide bandwidth, acting as entropy sources, enable scalable stochastic decision-making and optimize decision processes in a sixteen-armed bandit scenario. Finally, we systematically investigate the effect of external environmental change and the number of slot machines on the performance. Experimental and simulation results demonstrate that the proposed four-channel system based on a single multimode laser outperforms conventional single-channel systems, offering enhanced scalability and performance. This approach holds promise for scalable and robust applications in parallel decision-making and advanced reinforcement learning.