Android users know that, to type a word, the user can simply swipe a path over their digital keyboard and a word will be generated, automatically. This works sometimes, but there are misfires since a given swipe path can generate many combinations of words. Also, a user's swipe path can sometimes miss the intended letters on the keypad.
Samsung's patent application - US20180348891A1 - leverages deep learning (e.g., a trained neural network) to generate the user's word, based on a swipe input. In the simple example below, a user swipes over letters 'b', 'o', and 'y' on a digital keyboard. The system leverages deep learning to generate the word 'boy'.
In some cases, the neural network can be refined over time. For example, if a user is trying to generate the word 'boy' but swipes over 'b' 'p' 'y' instead, the system can recommend 'boy', 'buy', and other selections. The user can select 'boy' to clarify the intended word. This selection can be used as input to further train the neural network.
The next time that a user accidentally swipes over 'b' 'p' 'y', the system can automatically generate the word 'boy'. Thus, the word generator can 'learn' over time and improve accuracy.