Principles of Quantum Reinforcement Learning

Quantum computation is at the forefront of technological advancements, it exploits the principles of quantum mechanics to perform complex calculations by manipulating quantum bits (qubits). Unlike classical bits, qubits can exist in superposition states, allowing exponentially fewer computational steps than the most efficient classical algorithm for certain problems. Quantum logic operations applied to qubits are called quantum gates. Quantum gates are unitary transformations that operate on qubits in a quantum system. A sequence of quantum gates forms a quantum circuit. Classical reinforcement learning (RL) is a branch of machine learning where an agent learns to make decisions through interaction with an environment. The agent takes actions, receives feedbacks in the form of rewards or penalties, and aims to learn an optimal strategy, called policy, to maximize cumulative rewards over time. Quantum reinforcement learning is the intersection between reinforcement learning and quantum computing, and arises from the need to address challenges associated with quantum computations, such as the efficient preparation of quantum circuits. In this thesis, it is presented an overview of the employment of quantum data within classical RL algorithms to explore problem-solving in quantum contexts. This involves modeling complex quantum environments and seeking optimal policies within quantum settings. In particular, this work is focused on how to train an agent to design a quantum circuit, which transforms an initial state vector associated to a quantum system into a target state of interest, minimizing the number of quantum gates. For example to prepare entangled states of N qubits, such as a |00...0> + b |11...1>, starting from the initial state |00...0>. Designing optimal quantum circuits by minimizing the number of quantum gates represents a key aspect in quantum computing.