Q$7$ : Answer the following questions briefly:

$7\mathrm{.}1-$ What is the vanishing$/$exploding gradient problem in Recurrent Neural Networks?

$7\mathrm{.}2-$ For a problem like anomaly$-$detection, what kind of an architecture$/$model would you use?

Why?

$7\mathrm{.}3-$ What are the $2$ advantages of using Transformer models compared to LSTM and RNN$?$

$7\mathrm{.}4-$ Why do we use word$2$vec before building an NLP model? Can't we feed words$/$sentences

directly to our deep learning model? Explain briefly.

$7\mathrm{.}5-$ Assume we have negative and positive values in our training data. Shall we use relu

activation function for such data? Explain briefly.