Yonsei Univ. 23-1 STA4119 Final Project
Team [KimNaParkIm]
We present Quantum Bayesian Neural Networks (QBNN), which combine quantum computing and Bayesian inference to achieve superior modeling performance. QBNN represents the neural network's weights and biases using quantum bits (qubits) and employs quantum gate operations for parallel processing and measurements of quantum states. These quantum computing characteristics enable efficient handling of large data volumes and offer enhanced problem-solving capabilities in higher dimensions.
Moreover, QBNN leverages Bayesian inference principles to enhance uncertainty modeling. By combining prior distributions for weights and biases with likelihoods for the data, QBNN computes the posterior distribution and effectively handles uncertainty in the model. This approach improves the reliability of predictions by utilizing the results of Bayesian inference.
This experiment provides a comprehensive explanation of QBNN's implementation method and highlights its advantages. Additionally, we present experimental results that compare QBNN with traditional neural networks in image classification and regression tasks. The findings demonstrate that QBNN surpasses traditional neural networks in terms of accuracy and generalization ability.