Implemented a Neural Network from scratch using NumPy
Implementing algorithms from scratch is an effective way to gain a more profound knowledge and understanding of an algorithm. In this question, we implement a deep neural network from scratch with NumPy. For this purpose, the UTKFace dataset includes more than 23000 face images of people of different ages, races, and genders. In the first question, I train an artificial neural network for age prediction, then in the second part for race prediction, and last for gender prediction.
In the second question, I do calculations for a convolution layer. In the third question, I do a “Back propagation” operation for a simple convolution neural network that includes one convolution layer and one fully connected layer.