Individual Models for Touch Accuracy - towards probabilistic text input

I will present a flexible Machine Learning approach for learning user-specific touch input models to increase touch accuracy on mobile devices.  The model is based on flexible, non-parametric Gaussian Process regression and is learned using recorded touch inputs.  I will demonstrate that significant touch accuracy improvements can be obtained when either raw sensor data is used as an input or when the device's reported touch location is used as an input, with the latter marginally outperforming the former.  The learned offset functions are highly nonlinear and user-specific and that user-specific models outperform models trained on data pooled from several users.  Crucially, significant performance improvements can be obtained with a small (~200) number of training examples, easily obtained for a particular user through a calibration game or from keyboard entry data.