Cygames has 1 granted patent in 1 category: AI & Machine Learning (1). The patent covers technology for training game AI systems using replay logs from actual gameplay.
The approach involves weighting player history data, converting game states and actions into text representations, and generating order variants to predict subsequent moves.
Cygames discloses a single AI and machine learning patent that transforms replay logs into training data for transformer-based language models. The method assigns weights to history groups based on user information, then generates multiple reordered text samples of each game state in proportion to those weights. This text-order augmentation approach allows natural language models to learn from gameplay sequences in a way that earlier action-inference methods did not support. The system also incorporates randomized negative action pairs during training, which helps the model distinguish between actions a player actually selected and actions that were merely available at the time.
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.