請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/59827
標題: | 透過增強式學習攻擊區塊鏈拜占庭協議演算法:初步研究 Attacking blockchain Byzantine agreement algorithms with reinforcement learning: a preliminary study |
作者: | DUEN YIH TENG 鄧惇益 |
指導教授: | 鄭振牟(Chen-Mou Cheng) |
關鍵字: | 區塊鏈,拜占庭協議,分散式系統,資訊安全,增強式學習,評估分析工具, Blockchain,Byzantine Agreement,Distributed System,Security,Reinforcement Learning,Assessment Tool, |
出版年 : | 2020 |
學位: | 碩士 |
摘要: | 現在的區塊鏈越來越多,很多區塊鏈有用到拜占庭協議演算法,為了保證區塊鏈安全就需要拜占庭協議演算法是安全的,因此需要嚴格地數學證明來證明拜占庭協議演算法是安全的。雖然已經有很多拜占庭協議演算法被證明是安全的,但是有些還無法被證明是有效率的。由於不同的拜占庭行為會導致拜占庭協議演算法的執行時間不同,為了量化拜占庭協議演算法在最糟情況下的效率,我們需要一個模擬器和一個能讓演算法出現最糟情況的壞人,壞人藉由攻擊演算法使其出現最糟情況。再來現在的拜占庭協議演算法有數百個,在無法每個演算法效率都深入研究的情況下,我們提出一個機器,它可以在不深入研究拜占庭協議演算法地情況下,學習如何攻擊各種拜占庭協議演算法。如果一個拜占庭協議演算法在此機器的攻擊下,此演算法效率比預期的低,那此演算法就可以被認為不實際;反之,在機器的攻擊下,如果演算法效率還是很好,那此演算法會被認為很好用。藉由讓此機器攻擊各種拜占庭協議演算法,可以讓沒有研究背景且無法深入了解拜占庭協議演算法的使用者可以初步地知道哪些演算法是實際可用的。我們透過增強式學習攻擊拜占庭協議演算法來讓使用者可以對各種在不同壞人能力的情況下的拜占庭協議演算法效率有個初步結果。 There are more and more blockchains nowadays. Many blockchains use Byzantine agreement algorithms. To ensure the security of the blockchain, the Byzantine agreement algorithm is required to be secure. Therefore, strict mathematical proof is required to prove that the Byzantine agreement algorithm is secure. Although many Byzantine agreement algorithms have been proven to be secure, some have not been proven to be efficient. Because different Byzantine behaviors will lead to different execution times of Byzantine agreement algorithms, in order to quantify the performance of the Byzantine agreement algorithm in the worst case, we need a simulator and an adversary who can find the worst-case performance of the algorithm. The worst case is caused by the attack algorithm. Now, there are hundreds of Byzantine agreement algorithms. In the case where the performance of each algorithm cannot be studied in depth, we propose a machine that can learn how to perform various types of attacks without in-depth study of the Byzantine agreement algorithm. If a Byzantine agreement algorithm is under the attack of this machine, the performance of the algorithm is lower than expected, then the algorithm can be considered impractical; on the contrary, under the attack of the machine, if the performance of the algorithm is still very good, then this algorithm will be considered as practical. By using this machine to attack various Byzantine agreement algorithms, users who do not have a research background and can not have a deep understanding of Byzantine agreement algorithms can initially know which algorithms are still practical. We attack the Byzantine agreement algorithms with reinforcement learning so that users can have a preliminary result on the performance of the Byzantine agreement algorithms under various adversary's capabilities. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/59827 |
DOI: | 10.6342/NTU202003295 |
全文授權: | 有償授權 |
顯示於系所單位: | 電機工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-1308202016533100.pdf 目前未授權公開取用 | 4.04 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。