Accession Number:



Variety Preserved Instance Weighting and Prototype Selection for Probabilistic Multiple Scope Simulations

Descriptive Note:

Technical Report,22 Apr 2015,21 Apr 2017

Corporate Author:

Osaka University Osaka

Personal Author(s):

Report Date:


Pagination or Media Count:



Studies of probabilistic modeling and simulation focused on some special situations and behaviors, including rare events and scenarios, occurred in large scale and complex systems have been performed extensively. However, few studies have explored the modeling and simulation techniques to seamlessly cover multiple scopes including both major frequent and minor rare situations and behaviors embedded in a given large data set. A main obstacle to develop such techniques comes from the difficulty to capture the situations and the behaviors having highly contrasted probabilities in a unique model of the data distribution. Two technical issues must be addressed for overcoming this obstacle a weighting instances in a given large data set and b selecting prototypes from a given large data set. Particularly, the latter is for the modeling from massive data to which the thorough access is not feasible. A method to address these two issues preserves the variety of instance distributions of the data set and provide the basis of the seamless simulations over the multiple scopes. In the first year, we performed a mathematical analysis to derive the required conditions on our targeting method and designed a principle of the method which largely alleviate the obstacle by efficiently sampling the required prototypes with their appropriate weights. In the second year,we implemented the designed principle of the targeting method to an algorithm in computers and evaluated its generic performance preserving the variety of instance distributions of the data set in the selected prototypes. The mathematical analysis, the designed principle,the implemented algorithm and its performance evaluation presented in this report is the first work in worldwide for the seamless and comprehensive probabilistic simulations of the large scale and complex systems.

Subject Categories:

  • Numerical Mathematics

Distribution Statement: