In deep learning a lot of changes have come over the years and one such change is the use of Convolution Neural Network(CNN). CNN is a sub-domain of Artificial Neural Network (ANN), discovered by a postdoctoral researcher Yann LeCunn. In today’s time, deep learning is used in many industries with different applications like unmanned cars, news clustering, fraud news identification, processing high-level language, fraud detection, etc. Convolution neural networks are very useful in extracting distinct features of handwritten characters which makes them the natural choice for solving complex problems related to handwritten digit recognition. This paper aims to identify different optimization algorithms that can be used for handwritten recognition, evaluate those optimization techniques, and find the most accurate optimization technique.
Key words: convolution neural network, stochastic gradient descent, adam, Rmsprop (root mean square prop), adadelta, adagrad
|
