請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/68594
標題: | 殘差網路啟發的混合量子神經網路之研究 ResNet-Inspired Hybrid Quantum Neural Network |
作者: | Da-Young Chiu 邱大洋 |
指導教授: | 管希聖(Hsi-Sheng Goan) |
關鍵字: | 量子神經網路,量子機器學習,混合量子神經網路,殘差網路, Quantum Neural Network,Quantum Machine Learning,Hybrid Quantum Neural Network,ResNet,Residual Network, |
出版年 : | 2020 |
學位: | 碩士 |
摘要: | 在這篇碩士論文中,我提出了一個新的量子電路架構,而這個量子電路並非純量子電路,而是混合量子電路,因為在這個量子電腦尚未健全的年代(Noisy Intermediate-Scale Quantum Era)要使用深度非常深而且量子位元非常多的狀況短期內是不會實現的,而我所提出的『殘差網路啟發的混合量子神經網路(ResNet-Inspired Hybrid Quantum Neural Network )』的特性是深度淺所需的量子位元又少,而且精準度又比原來的純量子電路要高,再我看到『殘差網路(Residual Network)』後,深深的啟發了我,透過不一樣的編碼傳遞方式,讓我得到了新的量子電路,而我為了要展現出這個電路的架構的優點,我選了一個簡單的機器學習問題也就是『函數擬合(Function Fitting),用這個問題去展現這個量子電路的優點,而在這個問題下我比較了五個不一樣的模型對函數擬合的收斂狀況,發現在新的架構下『代價函式(Cost Function)』可以比原來的電路好至少兩個數量級,未來這個研究將可能可以成為各種量子機器學習的基石。 Inspired by the residual network (ResNet) structure in classical deep convolutional neural network, in this thesis we propose a new hybrid quantum neural network for machine learning tasks. Since in a noisy intermediate-scale quantum (NISQ) era, we can not run a deep quantum circuit on real quantum computers due to the lack of quantum error correction. And furthermore, we do not have a lot of qubits, and that is why I propose this new kind of quantum circuit. In the ResNet-inspired hybrid quantum neural network we propose the quantum circuit is shallower than the pure quantum circuit, and does not require many qubits,which may suitable for NISQ device. Furthermore, by creating a new way to encode and to transfer the data the ResNet-inspired hybrid quantum neural network could have better performance and accuracy on machine learning tasks. To demonstrate the advantage of ResNet-inspired hybrid quantum neural network, we pick a simple machine learning job which is function fitting. In this problem, we compare the performance and the convergence between 5 different quantum circuit models: pure quantum circuit,two different kinds of hybrid quantum circuit, and two different kinds of ResNet quantum circuits. We find that by using in the new ResNet circuit structure the accuracy or the value of the cost function is at least two orders of magnitude better than that by the pure quantum circuit and hybrid quantum circuit. The result of our study and the structure of ResNet quantum circuit could possibly serve as the building block of various quantum machine learning tasks. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/68594 |
DOI: | 10.6342/NTU202003705 |
全文授權: | 有償授權 |
顯示於系所單位: | 物理學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-1708202011405500.pdf 目前未授權公開取用 | 3.12 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。