As our project has ultimately required us to utilize Unreal Engine, I dug into some of the technology behind their newest launch with Unreal Engine 5. It heavily markets a few new features including Lumen (lighting processing), Nanite (graphics rendering), and Metahuman (high detail human faces and animations), which are all incredible powerful tools for the next generation of simulation and gaming. These are so impressive that I wanted to briefly discuss two of them here, Lumen and Nanite.

Lumen performs some serious magic and allows users to have effectively real-time ray-tracing from all the light sources (where light sources shoot near countless rays to determine what they can or cannot hit to determine shadows), including some very impressive radiosity (basically causing colors from nearby objects to reflect onto other objects).

Nanite interprets what the camera can or cannot see and intelligently stops the renderer from rendering polygons that the users perspective cannot even see (opposite side of walls, bottom of holes, etc.). This saves a huge amount of time for the renderer and increases performance significantly.

Now when I think of magic and intelligence in computer science, I can’t help but think of neural networks. Neural networks contain a variety of number of ‘layers’ where the model associates with the data in a particular way designated by the engineer. This could include things like pooling (identifying important features in data), convolution (randomly filtering inputs to see how it effects the output), and other such training optimizations. Now neural networks also often include what’s called a ‘hidden’ layer, where the model makes some associations within a black box. So when a neural network makes some incomprehensible connections, how could it be anything but magic?

So when technologies are discussed with these groundbreaking changes I often wonder what kind of methodologies did they use to determine these algorithms and processes, and how did machine learning impact these? While I imagine these are more closely guarded trade secrets, I feel we can get some insights from academia with the help of one of the greatest channels on YouTube, Two Minute Papers.