GDC Festival of Gaming
5 HN recommenders · recommended in 9 months, 2021–2025
1:05:25 · 35K views · Jun 18, 2020
Valve's John McDonald presents at GDC 2018 on how Valve uses deep learning (VACnet) to detect cheaters in Counter-Strike: Global Offensive. He covers training on human-reviewed replays and how model output is confirmed by manual review.
The go-to link in any anti-cheat thread for a concrete, well-delivered account of machine-learning cheat detection in production.
11 comments from Hacker News
Woah, cool to see a bit of how it all works. Seems similar to basic antivirus signature checking? As a side note, a few years back, they had a GDC talk about a (then-new) machine learning based approach to catching cheaters. Regardless of how it worked out, I found it fascinating. The presenter delivered the problem and solution approach so well I'd still like to work on something like this some time. I think they're still updating this for CS2, so it would be really cool to see how far it's gotten now. this video
Valve has been at this for a long time... They started with a system that lets players review games and try to understand if a player is a cheat or not. Games are anonymized, and you get to rewatch the replay to try to make a determination They can then use that data to train their NN to help detect which players / games should be fed into this manual review process I assume now the product is probably good enough that it's ranking reviewers too to determine how good reviewers are at detecting if there's cheats or not. It's pure genius Here's a GDC talk you may find interesting: > In this 2018 GDC session, Valve's John McDonald discusses how Valve has utilized Deep Learning to combat cheating in Counter-Strike: Global Offensive. this video
From what I know, VAC (Valve Anti Cheat) just looks for processes running on the system and detects injections into CSs memory. Then, for CS:GO specifically there's Overwatch, in which players look at other reported players' gameplay to determine whether they were cheating, and VacNET which is a machine learning system trained on the data from Overwatch to detect aimbots that way. There's a really good talk that someone from valve gave about 3 years ago[0]. The bigger problem is that even with input recognition, one of the biggest problems are wallhacks, meaning you can see other players through walls which is an advantage that's almost as large as aimbotting in tactical shooters like CS. [0]: this video
Or server-side automated review using machine learning models, which is what Valve is trying with "vacnet" (with then manual review to confirm the model's results are accurate - something most tech companies should learn from) Great talk on it here: this video
Unfortunately the Overwatch system has been disabled for years, and we currently have no reason to think it will be reopened for CS2. I take this opportunity to share this great talk about Valve's usage of deep learning to fight cheating in CS:GO: this video