Recurrent Neural Networks
- 91 Downloads
A feedforward fully-connected neural network cannot be used successfully for modeling sequences of data. A few basic reasons are the following: they cannot handle variable-length input sequences, do not share parameters, cannot track long-term dependencies, and cannot maintain information about the order of input data. Hence there is a need for a neural architecture that can handle successfully all the previous requirements.