Kinerja Deep Convolutional Network untuk Pengenalan Aksara Pallawa

This research trained Deep Convolutional Networks(ConvNets)  to classify hand-written Pallava alphabet.    The  Deep ConvNets  architecture consists of two  convolutional layers, each  followed by maxpooling layer,    two  Fully-Connected layers.  It had  442.602  parameters.  This model classified  660 images of hand-written Pallava alphabet into 33 diferent classes.  To make training faster, this research  used GPU implementation with 384  CUDA  cores. Two different techniques were implemented, Stochastic Gradient Descent  (SGD)  and Adaptive Gradient, each trained with 10, 20, 30 and 40 epoch. The best accuracy was 67,5%, achieved by the model with SGD technique trained at 30 epoch. Keywords: Deep ConvNets, Pallava, GPU, SGD