[CS231n] 1. Data-Driven Approach

This post has been written based on CS231n lecture.

Data-Driven Approach

To simply explain Data-Driven Approach, it is an approach to train a model with abundant data.

Before this approach, there were many attempts to create classifier models.

http://cs231n.stanford.edu/slides/2017/cs231n_2017_lecture2.pdf

The picture above well explains what was attempted:

1. Specify the appearance of a cat (“Has two ears, two eyes, how the nose looks like… etc)
2. Implement those characters as edges.
3. Implement the edges as algorithms.
4. Test

This sounds quite grinding. This edge method have several problems.

1. Different classes that has similar edges might also be classified as cats.
2. If the class changes, you need to implement the whole new algorithms.

To supplement these issues, the data-driven approach is adopted. It enables the model to train itself by data, making it needless to implement unnecessary-grinding works.