Resource Management for Chained Binary Classifiers

Networks of classifiers are capturing the attention of system and algorithmic researchers because they offer improved accuracy over single model classifiers, can be distributed over a network of servers for improved scalability, and can be adapted to available system resources. However, resource management for a network of classifiers, each with a range of operating points, is challenging because limiting one resource (e.g. CPU) may decrease the accuracy for that classifier, and increase other resource requirements (e.g. input bandwidth for downstream classifiers). In this paper we formulate the problem of optimal resource allocation for a chain of binary classifiers under generic resource constraints. We formally define a performance measure for the output of a chain of classifiers. We present our results for state-of-the-art classifiers operating on telephony data and offer interesting future directions based on these results.