[Paper Review] QuClassi: A Hybrid Deep Neural Network Architecture based on Quantum State Fidelity

Samuel A. Stein, Betis Baheri+6 more|arXiv (Cornell University)|Mar 21, 2021

Quantum Computing Algorithms and Architecture38 references16 citations

TL;DR

QuClassi proposes a hybrid quantum-classical deep neural network architecture that leverages quantum state fidelity for efficient binary and multi-class classification. By encoding data with fewer qubits and optimizing via a quantum differentiation function, QuClassi achieves up to 203.00% higher accuracy than QuantumFlow on multi-class tasks and reduces parameters by 97.37% compared to classical networks, outperforming state-of-the-art quantum models on real quantum hardware including IBM-Q and IonQ platforms.

ABSTRACT

In the past decade, remarkable progress has been achieved in deep learning related systems and applications. In the post Moore's Law era, however, the limit of semiconductor fabrication technology along with the increasing data size have slowed down the development of learning algorithms. In parallel, the fast development of quantum computing has pushed it to the new ear. Google illustrates quantum supremacy by completing a specific task (random sampling problem), in 200 seconds, which is impracticable for the largest classical computers. Due to the limitless potential, quantum based learning is an area of interest, in hopes that certain systems might offer a quantum speedup. In this work, we propose a novel architecture QuClassi, a quantum neural network for both binary and multi-class classification. Powered by a quantum differentiation function along with a hybrid quantum-classic design, QuClassi encodes the data with a reduced number of qubits and generates the quantum circuit, pushing it to the quantum platform for the best states, iteratively. We conduct intensive experiments on both the simulator and IBM-Q quantum platform. The evaluation results demonstrate that QuClassi is able to outperform the state-of-the-art quantum-based solutions, Tensorflow-Quantum and QuantumFlow by up to 53.75% and 203.00% for binary and multi-class classifications. When comparing to traditional deep neural networks, QuClassi achieves a comparable performance with 97.37% fewer parameters.

Motivation & Objective

• To address the limitations of existing quantum machine learning models in multi-class classification, particularly high qubit and parameter requirements.
• To develop a scalable quantum neural network architecture that maintains high accuracy with reduced qubit usage and efficient training.
• To demonstrate superior performance on real quantum processors, including IBM-Q and IonQ, compared to state-of-the-art quantum and classical models.
• To explore the feasibility of quantum speedup in deep learning by integrating quantum state fidelity into a hybrid quantum-classical framework.
• To provide a generalizable architecture that enables high-accuracy classification on 10-class datasets using near-term quantum hardware.

Proposed method

• QuClassi employs a hybrid quantum-classical architecture where classical data is encoded into quantum states using a parameterized quantum circuit with reduced qubit count.
• The model utilizes a quantum differentiation function to compute gradients and optimize parameters through backpropagation on the quantum hardware.
• Quantum state fidelity is used as a metric to evaluate and refine the quantum state output, ensuring optimal classification performance.
• The architecture is trained iteratively on quantum simulators and then deployed on real quantum platforms, including IBM-Q and IonQ, via Microsoft Azure Quantum.
• The method supports both binary and multi-class classification by extending the quantum circuit structure and loss function accordingly.
• Experiments use 8000 shots per epoch to measure circuit loss and ensure stability across noisy intermediate-scale quantum (NISQ) devices.

Experimental results

Research questions

• RQ1 Can a quantum neural network architecture achieve high accuracy in multi-class classification with significantly fewer qubits and parameters than classical models?
• RQ2 How does the use of quantum state fidelity improve classification performance in hybrid quantum-classical deep learning systems?
• RQ3 To what extent can QuClassi outperform existing quantum machine learning frameworks like TensorFlow Quantum and QuantumFlow on real quantum hardware?
• RQ4 What is the impact of hardware-specific noise and topology on the performance of quantum neural networks in practical deployment?
• RQ5 Can a quantum model achieve competitive accuracy on complex datasets like MNIST with 10-class classification using near-term quantum devices?

Key findings

• QuClassi outperforms TensorFlow Quantum by up to 53.75% in binary classification accuracy on real quantum hardware and simulators.
• On multi-class classification, QuClassi achieves a 203.00% performance improvement over QuantumFlow, demonstrating its superiority in complex tasks.
• The model achieves 96.15% accuracy on the Iris dataset using IBM-Q London, closely matching simulator results and showing robustness to hardware noise.
• On the 4-dimensional MNIST dataset, QuClassi-S achieves 95.98% accuracy on IBM-Q Rome, with only a 0.2% deviation from simulation, indicating high fidelity in real-world deployment.
• IonQ’s trapped-ion platform achieved 80.00% accuracy on a 3,6 configuration, outperforming IBM-Q Cairo (72.00%) due to lower circuit depth and no need for SWAP gates.
• QuClassi reduces model parameters by up to 97.37% compared to classical deep neural networks while maintaining comparable performance, highlighting its efficiency.