Finding a generic neural network architecture for image classification

Apekshit Jotwani
Rishabh Arora

ABSTRACT

Recently, neural networks and deep learning have become quite popular,
especially in the area of Image classification where each hidden layer
can identify a certain feature of the image. There are several
different network types like Convolutional neural network (CNN), CNN
with autoencoders and decoders, ImageNet, RESNet, AlexNet [7],
Recurrent networks like LSTM etc., which can be used for Image clas-
sification. Using popular Image datasets like MNIST and CIFAR-10, we
intend to find whether there is a particular network type which gives
results near to the best classification scores today for these
popular datasets. This network can be a great initial choice for most
Image classification problems encountered. We found that the network
architecture AlexNet is a good choice as a baseline model for image
classification tasks.