The simulation animates a name being fed into a recurrent cell one character at a time, showing the hidden-state vector morph with each new letter until the final state is mapped to a predicted language of origin.
Type or select a surname to feed into the network, then step or play through the characters one at a time to watch the hidden state update, and check the output panel for the predicted origin and its confidence across languages.
Name input/selector, step-through and play/pause controls, playback speed
In the original PyTorch tutorial this design is based on, the network correctly learns that endings like '-ez', '-ov', and '-ski' are strong giveaways of origin, purely from being shown thousands of labeled examples, without ever being taught a single explicit spelling rule.
The simulation animates a name being fed into a recurrent cell one character at a time, showing the hidden-state vector morph with each new letter until the final state is mapped to a predicted language of origin.
The simulation animates a name being fed into a recurrent cell one character at a time, showing the hidden-state vector morph with each new letter until the final state is mapped to a predicted language of origin.
Type or select a surname to feed into the network, then step or play through the characters one at a time to watch the hidden state update, and check the output panel for the predicted origin and its confidence across languages.
In the original PyTorch tutorial this design is based on, the network correctly learns that endings like '-ez', '-ov', and '-ski' are strong giveaways of origin, purely from being shown thousands of labeled examples, without ever being taught a single explicit spelling rule.