LSTM (Long Short-Term Memory)

LSTMs are another variant of Recurrent Neural networks (RNNs) which have the capability of learning long term dependencies, specifically in the sequence prediction kind of problem. It has feedback connections i.e.; it is capable of processing the entire sequence of the data apart from single data points such as images. It can be said as an incremented version of RNN which shows outstanding performance on a large variety of problems.

Advantages of LSTM over RNN

suppose there is a task to modify certain information in calendar, in that case RNN completely changes the existing data by applying a function. Where, LSTM makes small modification on the data by simple addition and multiplication that flow through cell states. This is an example of LSTM, how it forgot and remembers things selectively, which makes it as an improvement over RNN

Applications of Feed Forward Neural Network:

• o Language modelling
• o Machine translation
• o Handwriting recognition
• o Image captioning
• o Image generation using attention models
• o Question answering
• o Video-to-text conversion
• o Polymorphic music modelling
• o Speech synthesis
• o Protein secondary structure prediction

Bidirectional LSTM

Bidirectional LSTMs is an upgrade over LSTMs, each training sequence is presented forward and backward to separate recurrent nets. Both sequences are connected to the same output layer, Bidirectional LSTMs have complete information about every point in a given sequence, everything before and after it.