基於古典與量子卷積神經網絡的時間序列圖像分析

Time Series Image Analysis by Classical and Quantum Convolutional Neural Networks

周琪(Zhou Qi)

指導教授 ： 廖四郎

國立政治大學/商學院/金融學系/碩士(2020年)

https://doi.org/10.6814/NCCU202001059

摘要

時間序列是一維數據，但我們可以將其轉換成二維矩陣，最終圖像化後成三維張量。圖像化時間序列的映射需要能夠保留時間依賴性等時間序列重要特徵。卷積神經網絡是深度學習中一種重要的神經網絡，在計算機視覺領域有著非常多的應用，其對視覺信息的處理能力非常突出。所以我們將時間序列圖像和卷積神經網絡結合，研究從高維數據上提取數據特徵的可行性與能力，並最終用來進行時間序列圖像分類和未來趨勢預測。研究結果顯示，卷積神經網絡在時間序列圖像分類和未來趨勢預測上有良好表現，其策略表現可以超越基準線並獲得正的累積收益。另外，我們也對量子卷積神經網絡做了初步探討，我們闡述了相關理論，同時模擬構建了一個量子卷積神經網絡模型。研究結果顯示，量子卷積神經網絡是可行有效的，且在時間序列圖像分類和未來趨勢預測上能夠達到古典卷積神經網絡的水準。量子卷積神經網絡具有很大潛力且值得被研究。

關鍵字

時間序列圖像化 ； 古典卷積神經網路 ； 量子卷積神經網絡 ； 圖像分類 ； 趨勢預測

並列摘要

Time series is a one dimensional data, but we can transform it to be a two dimensional matrix, and finally obtain a time series image, which is a three dimensional tensor. A map of imaging time series should preserve some important properties of the time series such as the temporal dependency. Convolutional neural network is a kind of neural networks in deep learning, it is widely applied in the field of computer vision for its strong visual information processing ability. Therefore, we combine the time series image and the convolutional neural network together to research the feasibility and the ability of extracting features from the high dimensional data, and use the model to do the time series image classification and the future trend prediction. The result shows that the convolutional neural network has a good performance on time series image classification and future trend prediction. It could beat the baseline and has a positive cumulative return. In addition, we do a preliminary research on the quantum convolutional neural network. We describe the relative theory and simulate a quantum convolutional neural network model. The result shows that the quantum convolutional neural network is feasible and could reach the similar level of the classical convolutional neural network on both time series image classification and future trend prediction. Quantum convolutional neural network is potential and deserves to be studied.

並列關鍵字

Imaging time series ； Classical convolutional neural network ； Quantum convolutional neural network ； Image classification ； Trend prediction

參考文獻

[1] Cong, I., Choi, S., & Lukin, M. D. (2019). Quantum Convolutional Neural Networks. Nature Physics, 15, 1273-1278.

[2] Eckman, J. P., Kamphorst, S. O., & Ruelle, D. (1987). Recurrence Plots of Dynamical Systems. Europhysics Letters, 4 (91), 973-977.

[3] Erhan, D., Bengio, Y., Courville, A., Manzagol, P. A., & Vincent, P. (2010). Why Does Unsupervised Pre-training Help Deep Learning? Journal of Machine Learning Research, 11, 625-660.

[4] Heaton, J. B., Polson, N. G., & Witte, J. H. (2016). Deep Learning for Finance: Deep Portfolios. Applied Stochastic Models in Business and Industry, 33, 3-12.

[5] Heaton, J. B., Polson, N. G., & Witte, J. H. (2016). Deep Portfolio Theory. arXiv:1605.07230.