EA has applied for a patent utilizing deep-learning technology to improve automated match-making for players across all of its games. Using a neural network to analyze player skill and match outcomes should result in more satisfying matches for players. Different EA titles have tried various approaches to solve the problem of online match-making, from utilizing the classic competitive Chess ELO system to the Engagement Optimized Match Making System which was designed to keep gamers playing for longer online sessions. Both of those methods utilized in the past have their drawbacks which will ideally not be present in a neural network powered deep learning system.
Deep learning networks have become more popular in the last few years now that computer processing power is catching up to what was previously seen as science fiction. EA previously filed a patent for an automated coaching system designed to help players improve across all of its online titles, not just their sports franchises. Between the multiple patents EA has filed their goal of providing players a better, healthier online experience utilizing powerful neural networks and A.I. Learning is coming into focus
According to the patent filed by EA, its neural network will analyze players in a prospective pool by looking at the overall player statistics and then searching for players with the best estimated individual skill ratings to create the ideal match up. Estimated player score, such as their KDA (Kills, Deaths, and Assists) in a FPS title or goals scored in FIFA, will be calculated based off of prior match results and then calculate each individual player score to get an estimated rating for the team of players during a match. In practice the process will be similar to the terrain generating neural network EA also patented with millions of mathematical calculations taking place behind the scenes at the same time.
Within the patent application, EA claims that the neural network will take into account further variables, including how each player behaviors on specific maps or with individual players in a sports title. For example if a player usually plays Pathfinder in Apex Legends but changes and decides to play as Crypto for a match, the neural network will find other players with similar predicted match outcomes closer to their expected behavior as Crypto, which should result in a more balanced match up. Ideally matches made with the proposed neural network would eliminate the problem with smurfs Apex Legends for example.
Creating fun, well-balanced online match ups has been a problem developers have been trying to solve since games had the capability. As deep learning becomes more and more viable the millions of mathematical calculations needed to quantify human behavior should ideally result in a better online environment for players of all skill levels. Neural networks can already re-create Grand Theft Auto 5 so they should be able to detect players attempting to manipulate online matchmaking for easy wins. Deep learning neural networks may finally be able to solve one of gaming's longest running issues.