Kannada-Mnist Classification Using Skip CNN

In this paper, we tried out a new pipeline on the new handwritten digits dataset termed Kannada-MNIST for the Kannada script as well as a real-world handwritten dataset termed Dig-MNIST. The Kannada-MNIST was created to potentially serve as a direct drop-in replacement for the original MNIST dataset whiles Dig-MNIST serves as an out-ofdomain test dataset.We achieved a state-of-the-art result in challenging the baseline. Our model achieved 97. 53% accuracy on KannadaMNIST against 96. 8% and 85. 02% accuracy against 76. 1% on the Dig-MNIST dataset.