Spatial Model Personalization in Gboard

We introduce a framework for adapting virtual keyboard to individual user behavior by modifying Gaussian spatial model use personalized key center offset means and, optionally, learned covariances. Through numerous real-world studies, we determine the importance of training data quantity and weights, as well number clusters into which group keys avoid overfitting. While past research has shown potential this technique using artificially-simple keyboards games or fixed typing prompts, demonstrate effectiveness highly-tuned Gboard app with representative set users their real behaviors. Across variety top languages, achieve small-but-significant improvements in both speed decoder accuracy.