Constructing a High Performance, Dispersed, and Open Modular Platform for Financial Data Analysis and Risk Management (Ii) - Privacy Measures

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

MOST106-2221-E019-071

Translated Name/計畫中文名

建立高效能，分散，及開放模組之財務資料分析與風險管理平台 (II) - 隱私機制

Project Coordinator/計畫主持人

William Wei-Yuan Hsu

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Computer Science and Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=12484677

Year

2017

Start date/計畫起

01-08-2017

Expected Completion/計畫迄

31-07-2018

Bugetid/研究經費

602千元

ResearchField/研究領域

資訊科學--軟體

前置計畫中產生的FinancialCloud(雲端財務系統)是個可以依照客戶的喜好動態調整其中的畫面、演算法、資料來源的開放、分散式之模組化的服務平台。我們已經在這系統上建構可轉債評價系統，利變型保險評價系統，以及人身保險評價系統。另外也將匯率連動型人身保險(以一個幣值販售且由另一個幣值計費)模組串接起來。本計畫目標在於軟體方與硬體方面皆有提升FinancialCloud 的藍圖。隱私跟資安在FinancialCloud 上是不同的問題。使用者不單希望自己的資料被安全的送達運算服務端，但同時也希望可以維持資料內容的隱私。例如使用者將自己資料送給某個服務提供者來計算，但也同時希望服務使用者在不知道資料內容的條件下，可以算出客戶所需要的資訊。本計畫內我們將使用同態加密(Homomorphic encryption)來達成。初步成果使用的是部份同態加密法，只能支援部份的運算，造成資料必須在使用者與服務提供者間來回的傳遞。效能上非常地浪費時間與網路頻寬。我們將研究是否有完整同態加密方法，來解決這效能問題，並能提供多元運算來支持更多的商品評價。前導研究中我們使用的是虛擬機器為主。但目前我們有取得與租借到足夠的實體機與計算結點，將可把FinancialCloud 部署在實體架構上。我們選擇了將RDBMS 資料庫轉移成noSQL 文件儲存體資料庫(document store model)以求效率上的提升。最後，會依照文獻內寫的理論，使用硬體來加速所需用的的同態加密法。我們可將演算法設計成支援Open-MP 平行化模式，甚至如果有可能取得足夠的 GPU，可將一些運算照著理論敘述，下放到GPU 進行運算。 In the previous project, we have brought forth the FinancialCloud, an open distributed framework based cloud computing that can host modularize financial services which may be easily integrated on demand to meet dynamic user requirements. Experiments by constructing convertible bond pricing services, interest rate sensitive and life insurances are successful over the FinacialCloud by interconnecting the modules. We have extended the capabilities of FinancialCloud to handle annuities and quanto-type insurances, i.e., policies sold in one currency and invested in another currency. This project will focus on both the algorithmic aspect and hardware aspect in improving the privacy of the FinancialCloud. Privacy problem is different from security problems in the FinancialCloud. Users not only want their data to be delivered securely, but also wish it to maintain private. For example, a user will want to deliver their data to some service provider for calculation, but do not wish the service provider know what they are calculating. In this project, we will deal with privacy problems using homomorphic encryption algorithms. Preliminary results use partial homomorphic encryption mechanism proves effective. However, pricing operations must go back and forth between users and service providers many times to complete, leading to a very time-consuming and bandwidth wasting process. We will research on delivering fully homomorphic encryption mechanism into the FinancialCloud, improving efficiency and also supporting more types of options pricing. Previous projects were conducted on virtual machines. We have allocated and borrowed enough physical devices to construct our FinancialCloud this year, deploying software and services throughout the infrastructure. Next, we will transfer all databases from the RDBMS model to the noSQL document store model. Finally, we will develop parallel algorithms and use the advantages of Open-MP, and if possible, acquire some GPU for CUDA operations. According to literature, these hardware optimizations may provide speed increase for the homomorphic encryption algorithms.

Keyword(s)

雲端計算
金融商品評價
利率
最佳投資策略
隱私計算
Cloud computing
Derivative pricing
Life insurance
Interest rates
Optimal investment
Privacy computation

DSpace CRIS

Constructing a High Performance, Dispersed, and Open Modular Platform for Financial Data Analysis and Risk Management (Ii) - Privacy Measures

Details

Description