Recurrent Neural Networks (RNNs) is a class of neural network where connection between states form a directed cycle. RNNs specilize in making use of sequential information. Unlike traditional neural networks that assume all input are independent of each other, RNNs keep a "memory" of what has been caculated so far by performing same tasks for every element of the input sequence, with the output being dependent on the result of previous computations, hence the "recurrent" in the name.
Here is an example of how RNN(sketch-RNN) completes the doodle by making use of information from intial human input. To play,
Sketch-RNN Demos is created by David Ha, Jonas Jongejan, Ian Johnson, trained using TensorFlow
Here is an example of how RNN(sketch-RNN) completes the doodle by making use of information from intial human input. To play,
- Choose from model menu the doodle you want to draw (e.g. "face")
- Draw first few strokes with your mouse.
Sketch-RNN Demos is created by David Ha, Jonas Jongejan, Ian Johnson, trained using TensorFlow