Quantum neural network benchmarking: Time series predictions with finance forecasting as a use case

The main goal of this project is to replicate the findings of a 2022 paper by Barclays that utilizes Parametrized Quantum Circuits (PQCs) to forecast financial time series (Emmanoulopoulos & Dimoska, 2022). By faithfully reproducing the work of Dimitrios Emmanoulopoulos and Sofija Dimoska, the project aims to establish a benchmark for PQCs in time series forecasting. This benchmark serves as a foundation for exploring variations in datasets, circuit architectures, and hyperparameters.
To achieve this, the project closely follows the methodology outlined by Emmanoulopoulos and Dimoska to attain similar outcomes using the same dataset, model architecture, and evaluation metrics. Additionally, an alternative circuit architecture is implemented on the same dataset and trained with identical hyperparameters. The resulting outcomes will be compared to the original architecture.
The code is designed with reusability and maintainability in mind, adopting an approach enabling traceability of the experiments with availability on GitHub (https://github.com/CERN-IT-INNOVATION). This approach facilitates the expansion of the benchmark to explore various applications of PQCs. Functions and circuits are organized into scripts and libraries, thoroughly documented, and centralized for ease of parameter modification. A Notebook demo is provided for illustrative purposes. The code is optimized for GPU utilization using the JAX machine learning framework from Google.
This report offers a comprehensive overview, summarizing the findings, analysis, and insights derived from the project's endeavors.