Quantum vs Classical ML: Benchmarking Image Recognition on MNIST
A recent investigation published on arXiv (2605.27923) evaluates both classical and quantum machine learning techniques for recognizing images, specifically using the MNIST dataset of handwritten digits. The analysis involves comparisons between Classical Support Vector Machine (CSVM) and Quantum Support Vector Machine (QSVM), along with Classical Convolutional Neural Network (CCNN) and Quantum Convolutional Neural Network (QCNN). It assesses factors such as classification accuracy, computational time, number of parameters, and memory needs across different feature dimensions and sample sizes, utilizing both CPU and GPU setups. The goal of this research is to determine if quantum computing provides any tangible benefits over classical approaches for tasks related to computer vision.
Key facts
- Study benchmarks classical and quantum ML models on MNIST dataset
- Models compared: CSVM, QSVM, CCNN, QCNN
- Metrics: accuracy, runtime, parameter count, memory
- Experiments vary feature dimensionality and sample size
- Execution on CPU and GPU environments
- Published on arXiv with ID 2605.27923
- Motivated by computational limits of classical models
- Quantum computing explored as new paradigm for image recognition
Entities
Institutions
- arXiv