``` Forecasting Time series tool through using Parameterised Quantum Circuits (PQCs) as Quantum Neural Networks (QNNs). Forecasting specifically time series signals with simulated quantum forward propagation ``` #### Performance Comparison vs Classical Bidirectional Long Short Term Memory (BiLSTM) - Small Noise Coefficient (up to 40% of the amplitude of the deterministic signal)- Similar performance with few parameters vs the classical with 1000s of parameters - Large Noise Coefficient --> Outperform BiLSTM #### QNN benefits: 1. Faster training than classical models with a fully quantum optimisation algorithm 2. more accurate with small number of model parameters for noisy signals with deterministic components > Quantum Continuous Optimisation Algorithms are key here to help with making the entire training process forward and back propagation quantum. #### Challenge for QNNs is the **volume of data** they could be trained upon. Moreover, in a realistic scenario, the entire end-to-end process for training a QNN, needs to be combined with a classical computer that would pre-process the data and prepare the input so it can be encoded to quantum arrays of qubits and after the readout qubit is measured, it would need to be scaled back to derive the prediction. ## Applications of Time Series Forecasting using Quantum Machine Learning - Volatility Prediction - Asset Pricing - Algorithmic Trading Quantum Anomaly Detection for "customer centric" fraud prediction by autoencoders and Quantum Generative Adversarial Networks Covering **one year of transactional activity** is a large data set with billions of transactions to cover all seasonal trends. #### Challenges - Low number of fraudulent transactions (unbalanced data) - Variety of Fraudulent Patterns https://arxiv.org/pdf/2202.00599.pdf #quantum