Flashcards
How convolutions, pooling, residual blocks, LSTMs, attention, transformers, autoencoders, GANs and diffusion models work, including output-size and parameter-count problems.
What three ideas make convolutional layers efficient for images?
Tap to flip
Local connectivity: each unit looks at a small patch
Weight sharing: the same filter slides over the whole image
Translation equivariance: a shifted input gives a shifted feature map
Tap to see the front