Deep Learning Hyper-Parameter Optimization for Video Analytics in Clouds

A system to perform video analytics is proposedusing a dynamically tuned convolutional network. Videos arefetched from cloud storage, pre-processed and a model forsupporting classification is developed on these video streamsusing cloud-based infrastructure. A key focus in this work ison tuning hyper-parameters associated with the deep learningalgorithm used to construct the model. We further proposean automatic video object classification pipeline to validatethe system. The mathematical model used to support hyper-parameter tuning improves performance of the proposed pipeline,and outcomes of various parameters on system’s performance iscompared. Subsequently, the parameters that contribute towardsthe most optimal performance are selected for the video objectclassification pipeline. Our experiment-based validation revealsan accuracy and precision of 97% and 96% respectively. Thesystem proved to be scalable, robust and customizable for avariety of different applications.