Asymptotic Analysis of Convolutional Neural Network with Minimization of Computational Resources for Real Time Short Video Applications

Abstract

Real time deep learning on video streams made a challenging task for video analysis on the edge devices. A lot of research revolving deep neural networks which solve computational problems very well and there is also improvements in computational power. That helps move the images and train huge networks and neural networks is trying to solve or video analysis trying to solve is getting an understanding of the video doing video analysis for video understanding understand what we see in the video. CNN are employed with vision algorithms, where the image is convolved iteratively by many of the kernelsto extract the features behind it. The paper approach deals with using adaptive recognizer for the wide class of system with non-linearity functions in variation environment. The standard online algorithm used is gradient descent for adjusting weights of the convolutational neural network. The learning set is assigned applying bounded condition to infinite. To analyze the input variation variables with its limiting actions, asymptotic analyzers have been used. The limiting conditions assure the convergence of these algorithms globally in the nonlinear structure. The main factor is that not necessary of using penalty term to achieve sequence of the weights with the boundless. Overall computational resources aimed at minimizing energy consumption with memory through the derivation of activation function and back propagation with matrix multiplication is used as optimization algorithm which is same as feed forward pass algorithm.