Is it possible to upscale frames using deep learning?

So Ray Tracing is now possible thanks to a combination of task specific cores like RT and Tensor cores for denoising the initial image after a quick Ray Traced pass. Is it possible to use Tensor or Deeplearning to upscale frames instead of rendering them natively? Essentially upscaling something like 1080p image to 4K or even 8K without touching any of the raw GPU power involved in the rendering process? Since it can do something much more complicated like denoising, I am pretty sure this is not something out of the ordinary.

Thank y ou