How are LSTMs and Recurrent neural networks different ?

Question

Being a beginner in Deep Learning. I read about LSTMs and about Recurrect Neural Nets.
The basic idea of both is the same.
So how are LSTMs and Recurrent neural nets different ? and what are examples/application of those.

Samim Ekram · Accepted Answer

LSTM is a recurrent neural network actually hence you find them both similar . But unlike traditional RNNs, an LSTM network is well-suited to learn from experience to classify, process and predict time series when there are time lags of unknown size and bound between important events.LSTM networks have also been used for automatic speech recognition, and were a major component of a network that in 2013 achieved a record 17.7% phoneme error rate on the classic TIMIT naural speech dataset

You can also check CTC which is a better implementation on LSTM .

VISHNU SUBRAMANIAN · Answer

What is RNN : At times a neural network has to learn about previous/historical events in order to predict what comes next. Example : The computer has to predict when Modi returned from France , from the below sentence "Modi returned last night from france" . So in this example the Neural network needs to look back in the data by two words to answer when did Modi return. Vanilla Rnn works well when the number of steps you have to look back is minimal . Lets take another scenario , where we have the same information was available in the start of paragraph. So in such a case we need to look back by 1000s of words. And in such scenarios you hit the curse of vanishig gradients. That is the gradients which are back propogated across the network becomes too small or 0 . To overcome the curse of vanishing gradients , we look at only a few steps back , but give the network a memory/state which is used to carry useful information across multiple steps( in our example it could be a set of few words). LSTM does that for us. It uses a neural network and gates to decide what to keep in the state. If all this sounds scary , do not worry . Most of the complexity are hidden by most of the frameworks. An interesting link to know more about LSTM and RNN.Understanding-LSTM

VISHNU SUBRAMANIAN · Answer

What is RNN : At times a neural network has to learn about previous/historical events in order to predict what comes next. Example : The computer has to predict when Modi returned from France , from the below sentence "Modi returned last night from france" . So in this example the Neural network needs to look back in the data by two words to answer when did Modi return. Vanilla Rnn works well when the number of steps you have to look back is minimal . Lets take another scenario , where we have the same information was available in the start of paragraph. So in such a case we need to look back by 1000s of words. And in such scenarios you hit the curse of vanishig gradients. That is the gradients which are back propogated across the network becomes too small or 0 . To overcome the curse of vanishing gradients , we look at only a few steps back , but give the network a memory/state which is used to carry useful information across multiple steps( in our example it could be a set of few words). LSTM does that for us. It uses a neural network and gates to decide what to keep in the state. If all this sounds scary , do not worry . Most of the complexity are hidden by most of the frameworks. An interesting link to know more about LSTM and RNN.Understanding-LSTM

How are LSTMs and Recurrent neural networks different ?

Being a beginner in Deep Learning. I read about LSTMs and about Recurrect Neural Nets.The basic idea of both is the same.So how are LSTMs and Recurrent neural nets different ? and what are examples/application of those.

Being a beginner in Deep Learning. I read about LSTMs and about Recurrect Neural Nets.
The basic idea of both is the same.
So how are LSTMs and Recurrent neural nets different ? and what are examples/application of those.