At the past Games Developer’s Conference developer John McDonald opened the veil of secrecy over how Valve catches cheaters in his popular online shooter Counter-Strike: Global Offensive. To do this, the company had to build 1700 processors, which built the “most honest in the world” neural network.
I’m not sure that this is enough to process the data of all the players that can be more than 500,000 online at peak times. However, Valve has another farm with the same number of processor capacities. The neural network working on them isolates non-standard behavior and player performance.
McDonald argues that the system of complaints, which is used in the same Overwatch, has an efficiency of about 15%. While machine learning and neural networks can raise this figure to 90%.