EQNAS: Evolutionary Quantum Neural Architecture Search for Image Classification.

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Machine Learning

Background:

• Quantum neural networks (QNNs) offer advantages like speed and memory capacity for massive data.
• Classical neural networks struggle with training massive datasets.
• Current QNNs suffer from complex, artificially designed circuits and low classification precision.

Purpose of the Study:

• To propose an effective neural architecture search method, EQNAS, to enhance QNN performance.
• To overcome the limitations of manual QNN design in terms of circuit complexity and classification accuracy.

Main Methods:

• EQNAS employs an evolutionary approach: initializing quantum populations after image quantum encoding.
• It involves observing and evaluating population fitness, followed by iterative updates using quantum rotation gate updates, circuit construction, and interference crossover.
• The process continues until satisfactory fitness is achieved.

Main Results:

• Experiments demonstrate the feasibility and effectiveness of the EQNAS algorithm.
• Searched QNNs show significant improvements over original algorithms.
• Classification accuracy increased by 5.31% on MNIST and 4.52% on a warship dataset.
• Parameter reduction achieved: 21.88% on MNIST and 31.25% on the warship dataset.

Conclusions:

• EQNAS successfully enhances quantum neural network architecture search.
• The method leads to demonstrably better classification accuracy and reduced model complexity.
• EQNAS represents a significant advancement in optimizing QNNs for practical applications.