Neural Network algorithms, concepts and application developed from scratch in python using just numpy, scipy and matplotlib libraries.
-
Planar Data Classification
(keywords: binary/multiclass classification, He/Xavier initializations , gradient checking, adam optimization,
l2 regularization, decision boundaries)- Classification of planar data and plotting decision boundaries using a shallow neural n/w
-
MNIST Digit Recognition
(keywords: multiclass classification, normalization, softmax with better stablity , random minibatchs, learning rate finder, learning rate decay, gradient checking, adam optimization, l2 regularization, confusion matrix)- Handwritten Digit classification using the entire MNIST Dataset from http://yann.lecun.com/exdb/mnist/
- The classification was implemented using a 3 layer neural network with 784 input units, 30 units in the first hidden layer, 20 in the second hidden layer, 10 units representing the ten digits in the output layer.
- Accuracy of about 97.7 % was obtained on the test set