CNN Vocabulary in Modern Deep Learning
CNNs matter even when you are not building a full computer vision system.
They introduced and popularized ideas that appear across deep learning:
- local computation
- weight sharing
- feature maps
- hierarchical representations
- receptive fields
- channels
- efficient parameter use
Transformers are different from CNNs, but many comparisons use CNN language. For example, attention is often contrasted with local convolution because attention can connect distant positions directly.
Learning CNN vocabulary therefore helps you read modern architecture discussions with less friction.
The point is not to memorize historical architectures. The point is to understand the computational ideas that survived.
Enter 1 if feature maps and receptive fields are useful CNN vocabulary, or 2 if they are unrelated to convolution.
Compute it first, then check your number.
Enter 1 if this chapter is a full computer vision course, or 2 if it teaches reusable CNN ideas.
Compute it first, then check your number.