Photonic approaches to reinforcement learning are reviewed. We present single-photon-based and chaotic-laser-based decision makers to solve multi-armed bandit problems as well as entangled-photon-based social decision-making by utilizing the unique attributes of classical and quantum light.
Decision making in dynamically changing uncertain environments is one of the most important elements in information and communications technology, in applications ranging from resource assignment in data centers to wireless communications and search functions. Here, we review our research on the efficient physical realization or acceleration of decision making using photonics. The problem of interest is the multi-armed bandit (MAB) problem, where the purpose is to maximize the total rewards in unknown environments that involve difficult tradeoffs in the so-called exploration-exploitation dilemma. We describe the principle of solving MAB problems by utilizing the wave-particle duality of single photons in which the probabilistic attributes of single light quanta are employed for exploration. The principle is transformed into ultrafast laser chaos, where the chaotically oscillating irregular time series provides fast and scalable decision making abilities. The problem becomes even more difficult when multiple players are involved, in the so-called competitive MAB (CMAB) problem, where the expected value is related to maximizing the social benefit and ensuring fairness among individuals. We demonstrate that entangled photons can resolve the CMAB problem. Theoretical studies on photonic decision making are also reviewed, showing that six entities in total interact with each other according to an octahedral structure.
Recently, extensive studies on photonic reinforcement learning to accelerate the process of calculation by exploiting the physical nature of light have been conducted. Previous studies utilized quantum interference of photons to achieve collective decision-making without choice conflicts when solving the competitive multi-armed bandit problem, a fundamental example in reinforcement learning. However, the bandit problem deals with a static environment where the agent’s actions do not influence the reward probabilities. This study aims to extend the conventional approach to a more general type of parallel reinforcement learning targeting the grid world problem. Unlike the conventional approach, the proposed scheme deals with a dynamic environment where the reward changes because of the agent’s actions. A successful photonic reinforcement learning scheme requires both a photonic system that contributes to the quality of learning and a suitable algorithm. This study proposes a novel learning algorithm, a modified bandit Q-learning method, in view of a potential photonic implementation. Here, state–action pairs in the environment are regarded as slot machines in the context of the bandit problem and a change in Q -value is regarded as the reward of the bandit problem. We perform numerical simulations to validate the effectiveness of the bandit algorithm. In addition, we propose a parallel architecture in which multiple agents are indirectly connected through quantum interference of light and quantum principles ensure the conflict-free property of state–action pair selections among agents. We demonstrate that parallel reinforcement learning can be accelerated owing to conflict avoidance among multiple agents.
Artificial intelligence has permeated all aspects of our lives today. However, to make AI behave like real AI, the critical bottleneck lies in the speed of computing. Quantum computers employ the peculiar and unique properties of quantum states such as superposition, entanglement, and interference to process information in ways that classical computers cannot. As a new paradigm of computation, quantum computers are capable of performing tasks intractable for classical processors, thus providing a quantum leap in AI research and making the development of real AI a possibility. In this regard, quantum machine learning not only enhances the classical machine learning approach but more importantly it provides an avenue to explore new machine learning models that have no classical counterparts. The qubit-based quantum computers cannot naturally represent the continuous variables commonly used in machine learning, since the measurement outputs of qubit-based circuits are generally discrete. Therefore, a continuous-variable (CV) quantum architecture based on a photonic quantum computing model is selected for our study. In this work, we employ machine learning and optimization to create photonic quantum circuits that can solve the contextual multi-armed bandit problem, a problem in the domain of reinforcement learning, which demonstrates that quantum reinforcement learning algorithms can be learned by a quantum device.
The advantage of quantum computers over classical computers fuels the recent trend of developing machine learning algorithms on quantum computers, which can potentially lead to breakthroughs and new learning models in this area. The aim of our study is to explore deep quantum reinforcement learning (RL) on photonic quantum computers, which can process information stored in the quantum states of light. These quantum computers can naturally represent continuous variables, making them an ideal platform to create quantum versions of neural networks. Using quantum photonic circuits, we implement Q learning and actor-critic algorithms with multilayer quantum neural networks and test them in the grid world environment. Our experiments show that 1) these quantum algorithms can solve the RL problem and 2) compared to one layer, using three layer quantum networks improves the learning of both algorithms in terms of rewards collected. In summary, our findings suggest that having more layers in deep quantum RL can enhance the learning outcome.
Understanding and using photonic processes for intelligent functionalities, referred to as photonic intelligence, has recently attracted interest from a variety of fields, including postsilicon computing for artificial intelligence and decision making in the behavioral sciences. In a past study, we successfully used the wave–particle duality of single photons to solve the two-armed bandit problem, which constitutes one of the important foundations of decision making and reinforcement learning. In this paper, we propose and confirm a hierarchical architecture for single-photon-based decision making that verifies the scalability of the principle. Specifically, the four-armed bandit problem is solved given zero prior knowledge in a two-layer hierarchical architecture, where the polarization of single photons is autonomously adapted in order to effect adequate decision making. In the hierarchical structure, the notion of layer-dependent decisions emerges. The optimal solutions in the coarse layer and the fine layer, however, conflict with each other in some contradictory problems. We show that while what we call a tournament strategy resolves such contradictions, the probabilistic nature of single photons allows for the direct location of the optimal solution, even for contradictory problems, hence manifesting the exploration capability of single photons. This study provides insights into photon intelligence in hierarchical architectures for artificial intelligence as well as a novel aspect of photonic processes for intelligent functionalities.
Collective decision making is important for maximizing total benefits while preserving equality among individuals in the competitive multi-armed bandit (CMAB) problem, wherein multiple players try to gain higher rewards from multiple slot machines. The CMAB problem represents an essential aspect of applications such as resource management in social infrastructure. In a previous study, we theoretically and experimentally demonstrated that entangled photons can physically resolve the difficulty of the CMAB problem. This decision-making strategy completely avoids decision conflicts while ensuring equality. However, decision conflicts can sometimes be beneficial if they yield greater rewards than non-conflicting decisions, indicating that greedy actions may provide positive effects depending on the given environment. In this study, we demonstrate a mixed strategy of entangled- and correlated-photon-based decision-making so that total rewards can be enhanced when compared to the entangled-photon-only decision strategy. We show that an optimal mixture of entangled- and correlated-photon-based strategies exists depending on the dynamics of the reward environment as well as the difficulty of the given problem. This study paves the way for utilizing both quantum and classical aspects of photons in a mixed manner for decision making and provides yet another example of the supremacy of mixed strategies known in game theory, especially in evolutionary game theory.