Cygames received 1 granted patent this quarter in AI & Machine Learning.
The patent covers a method for training game AI using replay logs, where player history is weighted and game states and actions are converted into text representations. The system generates variations of action sequences to predict subsequent moves, enabling AI models to learn from recorded gameplay data.
Cygames filed 1 patent in AI and machine learning that rethinks how game AI learns from recorded player sessions. The method assigns importance scores to different chunks of replay history based on user metadata, then converts each moment of gameplay into text that can be fed to language models. To create training data, the system generates multiple shuffled versions of the same game state, with the number of copies proportional to how heavily that history is weighted. The approach also pairs correct player actions with deliberately incorrect alternatives, teaching the model to distinguish between what a player actually chose and what they merely could have chosen.
All data sourced from USPTO patent filings. Google Patents may take several weeks to index recent publications. If a link is unavailable, search for the patent number at USPTO Patent Public Search.