Member-only story
Deep Learning Fundamentals — Part Two
Welcome back, in Part one, we learned a couple of things:
- What is a Neural Network
- Problem of Classification
- Neural Network Structure
- Weights and Biases
if you didn’t read the first part, I suggest to read it before reading this part
We finished the first part with this question, why we need to create and train a web of classifiers for a task like classification, when potentially an individual classifier can do the job? In this part we going to talk about Pattern Complexity which can answer that.
Pattern Complexity
When you just need to analyze simple patterns in your data, using a basic tool like SVM or Logistic Regression is usually enough. However, if you have many input variables (e.g., 10 or more), neural networks tend to be superior.
However, for really complex patterns like human face detection, small neural networks with few layers may no longer be effective. This is because with more complex data, you need more processing power for each layer in the network and this increase is rapid and can get really costly in terms of traning and may not even give you good…
