Ever wondered how computers can create new images that look real?
Generative adversarial network
Ever wondered how computers can create new images that look real?
Imagine you're at a party and someone hands you a photo that looks like it was taken by a professional photographer. You're amazed by how realistic it looks, but you can't tell if it's real or fake.
A computer uses two parts to make fake photos. One part tries to spot fake photos, and the other tries to make them look real. They keep getting better at their jobs by competing against each other.
Example
The fake photo maker (generator) learns to make photos that look like real ones by seeing what the fake spotter (discriminator) likes and dislikes.
Remember this
Generative Adversarial Networks (GANs) create realistic fake photos by having two parts—the generator and the discriminator—compete to improve each other's skills.
Text adapted from Wikipedia, licensed under CC BY-SA 4.0.
Related concepts
data augmentation does for generalization: artificially expands training set
How can you teach a computer to see better?
DDIM does: deterministic sampling for faster generation with fewer steps
DDIM accelerates image generation by deterministically sampling intermediate steps
to use a CNN: for data with spatial structure like images or time series
Why can't we just feed all data into one big neural net?
message passing does in GNNs: each node aggregates features from its neighbors
Nodes in GNNs aggregate features from neighbors
the over-smoothing problem is in GNNs: deep GNNs make all node features converge
Over-smoothing in GNNs causes node features to converge
Pre-LN transformers are easier to train
Pre-LN transformers use residual connections, allowing gradients to flow more smoothly during backpropagation
Swipe through 100 ML concepts daily
Open Pocket Polymath