An ANN Model for image detection over the MNIST dataset from scratch using NumPy library.
Details of model:
No. of hidden layers: 2
Learning rate: 0.5
Optimizer implemented: adam
Regularizer: inverted dropout
Batch normalisation used: Yes
Normalized input features: Yes
Minibatch size: 1024
Epochs needed to converge: 100
Final training accuracy: 98.13 %
Test accuracy: 96.78 %