The European Quantum Technologies Conference (EQTC) 2024 is a significant event in the field of quantum technologies. This conference brings together experts, researchers, and industry professionals to discuss and showcase the latest advancements in quantum science and technology. It is an event run by Quantum Flagship, one of the most ambitious long-term Research and Innovation initiatives in Europe.
ParTec contributed to 2 poster sessions, one looking at the QML Framework with QKE algorithm for benchmarking QC, the other focussing on a QISKIT framework for QLSTM to predict real world time series data.
A summary of the sessions can be found below.
Poster Session: QML Framework with QKE algorithm for Benchmarking Quantum Computer in context of Application
In the era of NISQ (Noisy Intermediate-Scale Quantum) computing, hybrid quantum-classical workloads are likely to demonstrate the first significant advantages. Therefore, understanding the impact of noise on application results in such hybrid setups is crucial. Current benchmarks predominantly focus on assessing the quantum hardware stack alone, making it challenging to predict their effect on application outcomes.
To address this gap, we present a quantum machine learning benchmark framework using the quantum kernel estimation (QKE) algorithm within a standardized classification procedure employing a support vector classifier. QKE is well-suited for benchmarking applications due to its hybrid setup that spans the entire classical-quantum workflow, with the most computationally intensive parts executed on quantum hardware. It allows for easy adjustments to investigate the scale, quality, and speed of hybrid systems in relation to the accuracy of a trained model, using the widely used MNIST (Modified National Institute of Standards and Technology) dataset to achieve high accuracy without heavy reliance on data engineering.
Our research showcases results from a series of emulations on a pre-version of ParTec’s quantum workbench, varying the number of qubits, data sample sizes, quantum feature maps, and circuit depths. Initially, we establish a baseline with noise-free simulations on a fake backend, followed by the same emulations with noise models applied. This approach provides a comprehensive understanding of how noise impacts hybrid quantum-classical applications.
Poster Session: A QISKIT Framework for QLSTM to predict real world time series data
LSTM (Long Short-Term Memory) methods, introduced in the 1990s to address the vanishing gradient problem of recurrent neural networks, remain a valuable tool in deep learning and have recently been enhanced. LSTM methods are versatile, applicable in language modeling, machine translation, image captioning, handwriting generation, and image generation using attention models.
Building on initial quantum LSTM (qLSTM) approaches, we present a qLSTM framework integrated with PyTorch workflow in Qiskit. Our demonstration uses LSTM to predict time series data within a hybrid quantum-classical workflow, leveraging variational quantum circuits (VQAs) to represent the weights in the classical LSTM cell. The VQAs consist of an encoding layer (quantum feature map) to load classical data into the quantum feature space, and a variational layer (ansatz) with tunable parameters.
We showcase examples using the Aer state vector simulator for noise-free quantum computing simulations, as well as fake backends with noise models from current IBM Eagle devices. To demonstrate the functionality of the qLSTM framework, results are presented with toy data from a damped oscillator and real-world data used to create global warming stripes. This session highlights the potential and practicality of qLSTM in hybrid quantum-classical environments for various predictive modeling tasks.
If you are interested in learning more about the poster sessions, you can download the materials below:
Further insights
Stay tuned for more updates as ParTec continues to lead the charge in HPC, AI and Quantum Computing. You can find more insights and the latest news on our dedicated News & Insights page!
