Photonic decision-making for arbitrary-number-armed bandit problem utilizing parallel chaos generation.

Area of Science:

• * Chaos theory
• * Machine learning
• * Signal processing

Background:

• * Multi-armed bandit problems are common in decision-making scenarios.
• * Existing solutions may struggle with an arbitrary number of arms.
• * Chaotic signals offer a unique source of randomness.

Purpose of the Study:

• * To propose and demonstrate a novel scheme for solving any-number-armed bandit problems.
• * To utilize chaotic signals and the epsilon-greedy strategy for decision-making.
• * To investigate the impact of mapping rules and reward probabilities on performance.

Main Methods:

• * Generation of two parallel chaotic signals.
• * Processing signals via an 8-bit analog-to-digital converter (ADC) with 4 LSBs.
• * Employing the epsilon-greedy strategy for mapping sequences to arms.

Main Results:

• * Successful demonstration of decision-making for a 5-armed bandit problem.
• * Investigation of correction decision rate (CDR) for 4- to 7-armed problems.
• * Analysis of the influence of mapping rules and unknown reward probabilities.

Conclusions:

• * The proposed scheme offers a novel approach to solving arbitrary-number-armed bandit problems.
• * Chaotic signal generation and ADC processing provide uniform random sequences for decision-making.
• * The method shows promise for adaptive decision-making in complex environments.