Constructing Neural Networks From Scratch: Part 1

With the presence of popular deep learning frameworks such as TensorFlow, Keras, PyTorch, and other similar libraries, it has become much easier for a novice in the field to pick up the subject of neural networks at a faster pace. Although these frameworks provide you a path to solving the most complex computations within a few minutes, they don’t require you to understand the true core concepts and intuition behind all the requirements. If you have the knowledge of how a specific function works, and how exactly you can make use of that function within your code block, you will solve most problems without too much difficulty. However, for anyone to truly appreciate the concept of neural networks and understand the complete working procedure, it becomes essential to learn how these artificial neural networks work from scratch. How can these neural networks solve these complex problems?

Understanding how neural networks work and, subsequently, how they are constructed is a worthwhile endeavor for anybody interested in AI and deep learning. While we are restricting ourselves from using any kind of deep learning frameworks such as TensorFlow, Keras, or PyTorch, we will still make use of other useful libraries like NumPy for numerical matrix computations. With NumPy arrays, we can perform numerous complex computations that will mimic the effect of deep learning, and use that to build understanding of the procedural workflow of these neural networks. We will implement some neural networks designed to solve a fairly simplistic task with the help of these neural nets built from scratch.