mixed precision training does: forward in FP16, accumulate gradients in FP32

Ever wished your phone's camera could take stunning photos even in low light?

Imagine you're trying to take a great photo at night, but your phone struggles to capture details in dim lighting.

DLSS works like a smart filter that lets your phone's camera see better in the dark, then magically makes the photo clearer and more detailed.

Example

You snap a photo in low light, and instead of a blurry mess, you get a sharp, bright image as if it were daylight.

Remember this

DLSS uses a clever trick to process images at lower resolutions for speed, then reconstructs them at higher quality.

Text adapted from Wikipedia, licensed under CC BY-SA 4.0.

Related concepts

to standardize: when you need zero mean and unit variance for gradient-based optimization

Why do we need to make data uniform before training a model?

gradient clipping does: caps gradient norm to prevent exploding gradients

How do deep learning networks avoid getting stuck or going haywire during training?

gradient accumulation simulates larger batch sizes without more memory

Can you train a machine like you do with a computer?

DDIM does: deterministic sampling for faster generation with fewer steps

DDIM accelerates image generation by deterministically sampling intermediate steps

AdaGrad does: divides learning rate by sqrt of sum of squared gradients

How do we avoid overshooting in learning?

Reparameterization trick

Reparameterization trick enables differentiable sampling for VAE training

Swipe through 100 ML concepts daily