Features Added:
1. FeedForwardNetwork model which can be used to compose layers and train on data.
2. Layers
- Dense
- Dropout
3. Activation Functions
- ReLU
- Sigmoid
4. Objective Functions
- Stochastic Gradient Descent
- SGD with Momentum
5. Weight Initializers
- Zeros
- Random Uniform
6. Metrics
- Categorical Accuracy
7. Datasets
- MNIST handwritten digits
- Synthetic datasets (moons, spirals)