Cleaning Uncertain Streams by Parallelized Probabilistic Graphical Models

Abstract

Real-world applications generate uncertain streams due to unreliable equipments and/or data processing such as object identification. However, application context implies specific rules, which are critical in cleaning data and make them closer to the reality. In this paper, we propose a framework for cleaning uncertain streams by Parallelized Probabilistic Graphical Models (P²GM). Making full use of multi-core processing architecture, the system processes parallelized high-volume streams efficiently. With P²GM, users can define their own cleaning algorithms and generate specific parallelized systems. We implement a prototype of video surveillance based on P²GM, and demonstrate the quality and performance of our approaches experimentally.