Which statements about LSTM variants are true?

a$.$ The key idea of multidimensional LSTM $($Graves et al$.,2007)$ is to enable the network to process sequences consisting of vectors which stem from high$-$dimensional vector spaces.

b$.$ Convolutional LSTM $($Shi et al$.,2015)$ can process sequences of images, e$.$g$.,$ videos.

c$.$ Peephole connections $($Gers and Schmidhuber, $2000)$ provide gates with direct read access to the cell state.

d$.$ The Bidirectional LSTM $($Schuster et al$.,1997)$ has two layers. The second layer receives input from the first layer and reverses the processing direction.

e$.$ Stacked LSTM $($Graves et al$.,2013)$ stack LSTM layer by using one LSTM$\text{'}$s final cell state as the initial cell state for the next LSTM$.$