Member-only story
Using the right dimensions for your Neural Network
When building your Artificial Neural Networks, one of the frustrating bugs is getting the shape of the vectors and matrices wrong. This is a brief guide to understand the underlying math operations, and shed some light on the dimensions to pass in.
Updated — 8 August
This series has been simplified to a 3-part series.
Objective
If you are just embarking on Machine Learning, you would have come across basic models such as sequential networks. These Fully-Connected Neural Networks (FCNN) are perfect exercises to understand basic deep learning architectures before moving on to more complex architectures.
Beginners will find it easy to get started on this journey through high-level libraries such as Keras and TensorFlow, where technical details and mathematical operations are abstracted from you. The benefit is the ability to jump-start the learning journey without bogged down by the math, but it does introduce problems very quickly if you do not really know what you are doing.
If you are like me, you might have encountered situations where your code does not work the way you expected, and by…
