基於cifar-10的邊緣計算優勢和模擬

Abstract

現今物聯網需要的響應時間越來越短，而資料蒐集過程可能會包含私人數據，眾多的設備將會產生非常龐大的數據，雲端計算將無法處理未來將產生的龐大數據[1]，對於個人資料的保護和網路的傳輸成本也是一個問題。為了解決上述這些問題，可以採用邊緣計算將雲端的模型直接放在終端裝置上，如此不但可以解決個人資料傳輸保護的問題，網路傳輸以及雲端計算無法負荷的問題也得以解決，在響應時間上也可以得到最短的優化。 在本篇報告中，我設計一個參數量較多、較為龐大的模型用來模擬放在雲端的模型，然後再設計一個參數量較少、較為小的模型用來模擬放在終端裝置上的模型，訓練資料使用cifar-10資料集。 實驗結果驗證了邊緣計算在訓練模型和設計模型的時間上都比傳統雲端計算還要快，模型正確率也比傳統雲端計算還來的高。

Since some IoT management might require very short response time, some merge involved private data, and some might produce a large quantity of data which could be a heavy load for networks. Cloud computing is not expensive enough to support these Applications. [1] To solve these problems, I propose a method using edge computing. This method of edge computing can not only avoid the problem of protecting private data but also increase the effectiveness of network transmission, response time and cloud computing. In this paper, I design a model having more parameters and bigger size for simulating cloud computing and a model having less parameters and smaller size for simulating edge computing. I use cifart-10 dataset as training data for this experiment. The result of my experiment proves that edge computing is faster than traditional cloud computing in both training and design, and the model accuracy is also higher than traditional cloud computing.