Configuring topologies of distributed semantic concept classifiers for continuous multimedia stream processing

Abstract

Real-time multimedia semantic concept detection requires instant identification of a set of concepts in streaming video or images. However, the potentially high data volumes of multimedia content, and high complexity associated with individual concept detectors, have hindered its practical deployment. In this paper, we present a new online concept detection system deployed on top of a distributed stream mining system. It uses a tree-topology of classifiers that are constructed on a semantic hierarchy of concepts of interest. We introduce a novel methodology for configuring such cascaded classifier topologies under constraints on the available resources. In our approach, we configure individual classifiers with optimized operating points after jointly and explicitly considering the misclassification cost of each end-to-end class of interest in the tree, the system imposed resource constraints, and the confidence level of each object that is classified. We describe the implemented application, system, and optimization algorithms, and verify that significant improvement in terms of accuracy of classification can be achieved through our approach.

A framework for distributed multimedia stream mining systems using coalition-based foresighted strategies

“…Consider a stream mining application [7], which consists of several binary classifiers in a tree topology depicted in Fig. 1.…”

“…· Performance of CU: As in [7], performance of c i (c i ) is controlled by its tradeoff between probability of false alarm p …”

“…· Input and Output Rates: For c i andc i , the output stream rates (t i , g i ) and (t i ,ḡ i ) may be derived as [7]:…”

“…using a Support Vector Machine, and classifiers are organized into a semantic hierarchy of concepts. We compare the performance of myopic and foresighted strategies against a centralized approach [7].…”

“…In this paper, we focus on the classifier configuration problem for binary tree topologies, i.e., determining the optimal operating point (detection -false alarm tradeoff) for each classifier in the tree, in order to maximize the end-to-end classification performance. Previously, this problem has been modeled as as an optimization problem and centralized techniques such as Sequential Quadratic Programming (SQP) [7] have been used to solve it. However, a centralized solution suffers disadvantages in terms of having a single central point of control and associated failure, issues with scaling and adaptation as the topology grows, and not allowing large scale applications with capabilities distributed across multiple proprietary entities.…”