After the denoiser of raytracing images from Nvidia, here is a neural network approach from Intel to make game output photorealistic. Using the G-buffers as input is particularly clever.

Fascinating results. Could have implications both for neurosciences and machine learning.

Very interesting work on the use of sparse neural networks for faster inference. This is a good way to run originally larger models with less CPU and memory penalty than using the original model while retaining result quality.

Fascinating findings around artificial multimodal neurons. Especially interesting is how they encode information but also how they open the door to new attacks to the corresponding neural networks.