Reinforcement learning (RL) policy gradient methods, while effective, frequently encounter challenges related to instability and the sensitivity of their performance to the learning rate hyperparameter. This work introduces a novel approach leveraging quantum-inspired optimization techniques to address these issues. Specifically, we propose a quantum-inspired learning rate scheduler designed to dynamically adapt the update step size during policy gradient training. This scheduler, rooted in principles of quantum measurement and superposition, aims to provide a more robust and efficient learning process. The core claim of this paper is that the integration of this quantum-inspired learning rate scheduler significantly improves the stability and convergence speed of policy gradient methods. We present a theoretical framework and explore the potential of this approach, demonstrating its promise for enhancing RL performance. The proposed method offers a new direction in RL optimization, potentially bridging the gap between classical and quantum-inspired approaches.
Reinforcement learning is a growing field in AI with a lot of potential. Intelligent behavior is learned automatically through trial and error in interaction with the environment. However, this learning process is often costly. Using variational quantum circuits as function approximators potentially can reduce this cost. In order to implement this, we propose the quantum natural policy gradient (QNPG) algorithm - a second-order gradient-based routine that takes advantage of an efficient approximation of the quantum Fisher information matrix. We experimentally demonstrate that QNPG outperforms first-order based training on different Contextual Bandits environments regarding convergence speed and stability and moreover reduces the sample complexity. Furthermore, we provide evidence for the practical feasibility of our approach by training on a 12-qubit hardware device.
Variational Quantum Circuits are being used as versatile Quantum Machine Learning models. Some empirical results exhibit an advantage in supervised and generative learning tasks. However, when applied to Reinforcement Learning, less is known. In this work, we considered a Variational Quantum Circuit composed of a low-depth hardware-efficient ansatz as the parameterized policy of a Reinforcement Learning agent. We show that an
ε-approximation of the policy gradient can be obtained using a logarithmic number of samples concerning the total number of parameters. We empirically verify that such quantum models behave similarly or even outperform typical classical neural networks used in standard benchmarking environments and in quantum control, using only a fraction of the parameters. Moreover, we study the Barren Plateau phenomenon in quantum policy gradients using the Fisher Information Matrix spectrum.