• Valve wants to take a 'machine learning' approach to Counter-Strike anti-cheat
    12 replies, posted
[url]http://www.pcgamer.com/valve-wants-to-take-a-machine-learning-approach-to-counter-strike-anti-cheat[/url]
I want them to take at least some approach other than what they're currently doing, because it clearly doesn't work.
They have so many unfinished projects right now... (SteamOS, OpenVR, Steam Support and other stuff)
[QUOTE=Nicolas;51829861]They have so many unfinished projects right now... (SteamOS, OpenVR, Steam Support and other stuff)[/QUOTE] And im sure the people working on those still are working on those.
[QUOTE=Nicolas;51829861]They have so many unfinished projects right now... (SteamOS, OpenVR, Steam Support and other stuff)[/QUOTE] I wish people understood better that too many cooks also applies to programming as well and that you can't just expect them to throw all of their team at one project and expect it to work that and it's good to maintain a steady workflow for people who might need it within the company.
And people at valve just work on whatever they want to work on. Wheelie Desks Other development teams could take a hit if people decide to work on this project, but they must make the choice on their own if they believe that what they are working on currently is or is not more important with other opportunities within the compary
Sounds like what [url=https://www.youtube.com/watch?v=ghRYZJy95oo]HackCam promised to be[/url] back in 2004. It was bought out by an unknown company, and never heard from again (other than a developer on the forums once telling a highly respected member of the Australian CSS community to stop "shitting on my lawn").
I don't think more automation is what Valve needs right now. Or ever.
[QUOTE=WaLLy3K;51830102]Sounds like what [url=https://www.youtube.com/watch?v=ghRYZJy95oo]HackCam promised to be[/url] back in 2004. It was bought out by an unknown company, and never heard from again (other than a developer on the forums once telling a highly respected member of the Australian CSS community to stop "shitting on my lawn").[/QUOTE] could you imagine having to be the players in that test footage who didnt have hacks.
[QUOTE=WaLLy3K;51830102]Sounds like what [url=https://www.youtube.com/watch?v=ghRYZJy95oo]HackCam promised to be[/url] back in 2004. It was bought out by an unknown company, and never heard from again (other than a developer on the forums once telling a highly respected member of the Australian CSS community to stop "shitting on my lawn").[/QUOTE] Person running past double doors gets shot at (Above average-reaction) alright boys, time to pack up those double door strats, the deal is up.
This could prove to be very effective if implemented well. They have massive amounts of training data for a system like that thanks to the player-driven Overwatch system.
[QUOTE=WaLLy3K;51830102]Sounds like what [url=https://www.youtube.com/watch?v=ghRYZJy95oo]HackCam promised to be[/url] back in 2004. It was bought out by an unknown company, and never heard from again (other than a developer on the forums once telling a highly respected member of the Australian CSS community to stop "shitting on my lawn").[/QUOTE] We've progressed a lot in machine learning from 2004; for proof just look at Google's Knowledge Graph, and their [URL="https://cloud.google.com/vision/"]Vision API[/URL]. I doubt Valve is completely rolling their own custom solution here, they're like using something like TensorFlow, on a large provider such as GCP or AWS. [editline]16th February 2017[/editline] It's also worth noting that the slowest part of ML is training the neural net, that and having a large sample size can be hard to get. Since they're probably going to be using OW data for this, they'll have the sample size, and plenty of time to train the net offline on that training set. Once they have the trained net, they can have it weigh in on OW cases after-the-fact, eventually they'll have a refined enough set (that runs fast enough) and they'll be able to do it in near-realtime.
I'm surprised it took them so long to consider this, Antiviruses and other software could really use a neural AI and machine learning networks. I would also like to see AIs being implemented into graphics, for the ambient occlusion, correct fast antialiasing, and other similar things.
Sorry, you need to Log In to post a reply to this thread.