The author investigated the application of Long Short-Term Memory (LSTM) Recurrent Neural Networks (RNNs) to the task of signature verification. Traditional RNNs are capable of modeling dynamical systems with hidden states; they have been successfully applied to domains ranging from financial forecasting to control and speech recognition. This manuscript is the result of successfully applying on-line signature time series data to traditional LSTM, LSTM with forget gates and LSTM with peephole connections algorithms originally developed by S. Hochreiter and J. Schmidhuber. It can be clearly seen in this pattern classification problem that traditional LSTM RNNs outperform LSTMs with forget gates and peephole connections. The latter also outperform traditional RNNs which cannot seem to even learn this task due to the long-term dependency problem.
|Number of Pages||104|
|Book Type||Computer networking & communications|
|Country of Manufacture||India|
|Product Brand||LAP LAMBERT Academic Publishing|
|Product Packaging Info||Box|
|In The Box||1 Piece|
|Product First Available On ClickOnCare.com||2015-07-27 00:00:00|