Quantum-enhanced generalisation advantage for recurrent learners with minimal states

• Main Author: Prahara, Aurelio Jethro
• Other Authors: Gu Mile
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Theory of computation::Analysis of algorithms and problem complexity; Engineering::Computer science and engineering::Theory of computation::Computation by abstract devices; Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/166740
• Institution: Nanyang Technological University

Quantum Computers have demonstrated several speedups with Shor’s Algorithm and Grover’s Algorithm. However, there is still a lot of possible advantages that have not been explored. One of these avenues is in the realm of intelligent agents. The objective of this project is to explore the discretization of Recurrent Neural Networks (RNNs) into Finite State Machines (FSMs) through two main methods, ternary neurons and clustering. We demonstrated that these two methods produce the same FSM but however, for a simple communication game, it did not produce the minimum FSM. We also showed that for the game of Pong, it did produce a minimal FSM although there is no quantum advantage in this case.