Shivanshu Siyanwal

Hackathon · QML

Hybrid Quantum LSTM (QPoland Hackathon 2025)

Submission to the QPoland Global Quantum Hackathon 2025, Quantum in Finance track: a hybrid quantum LSTM that predicts closing prices on the test set.

Participant (team PlanckShift), QPoland Global Quantum Hackathon 2025, Quantum in Finance track, 18–26 October 2025. QML / PennyLane / Time series

Layman view

In plain words

A plain-language version is coming soon.

Expert view

The work, step by step

  1. 01

    Task

    Predict closing prices on the test set of the Quantum in Finance track, QPoland Global Quantum Hackathon 2025.

  2. 02

    Model

    Inputs are windows of 8 consecutive time steps of seven min–max-scaled features from the price data: open, high, low, volume, the high–low range, the high/low ratio and a 3-step moving average of volume. The target is the closing price.

    The LSTM's input, forget and output gates are classical. Its cell candidate comes from a 4-qubit circuit simulated with PennyLane: PCA reduces the current input and hidden state to four values, which are angle-embedded and passed through two basic entangler layers; the Pauli-Z expectation values, passed through tanh, form the candidate.

    Training uses Adam with weight decay, a Huber loss, dropout and early stopping, on a chronological 65/35 split into training and validation data.

  3. 03

    Repository

    The notebook, the trained weights, the training metrics and the test-set close-price predictions.

Contribution

My role

Team PlanckShift, which included Kshitij Dave. I contributed equally to the research, code development and testing.

More threads