For testing this very early and not final version of the qLSTM demonstrator, we recommend to clone the GitHub repository: https://github.com/drstki/qlstm-with-qiskit-and-pytorch/tree/main Install a virtual python environment and install the necessary Qiskit and ML frameworks on your local workstation. You can then execute the demonstrator notebook locally. https://github.com/drstki/qlstm-with-qiskit-and-pytorch/blob/main/qlstm_GlobalWarming_demonstrator.ipynb You find Qiskit installation instructions at https://github.com/drstki/qlstm-with-qiskit-and-pytorch/blob/main/qlstm_GlobalWarming_demonstrator.ipynb
About
We will build a quantum quantum Long Short-Term Memory (qLSTM) demonstrator with the quantum development framework Qiskit V1.x. LSTM can be used in different use cases, e.g. language modeling, machine translation (sequence to sequence learning), image captioning, handwriting generation, image generation using attention models. In our demonstrator qLSTM we use LSTM for predicting time series data. We are utilizing the integrated PyTorch workflow in Quskit to build a hybrid quantum-classical qLSTM program. We use variational quantum circuits (VQA) for representing the weights in the classical LSTM cell. The variational quantum circuits are built of 2 parts: An encoding layer (quantum feature map) to load classical data points into quantum feature space, and a variational layer (ansatz) which contains the tunable parameters. The demonstrator will use the Aer statevector simulator for noise free QC simulations. We will also provide fake backends which can be used with different noise-models to simulate qLSTM in a more realistic environment and to learn to deal the noise in quantum devices. We plan to provide also different data sources which can be used for demonstration purposes.
