Backpropagation through time $($BPTD$)$ is the standard gradient computation algorithm for FNNs$.$ However, there exist altematives, eg$.$ truncated BPTI and real$-$time recurrent learning $($RIRL$).$ Which statements about these algorithms are true.

a$.$ Truncated BPIT divides the input sequence into chunks and applies BPIT within these chumks but not across thern.

Trumeated BPTI gradients are different from BPIT gradients.

c$.$ Truncated BPIT has advantages over BPIT if the input sequences are short.

d$.$ RIRL is very fast and therefore often used in practice.

e$.$ The memory complexity of RIRL is $O(N^3),$ where $N$ denotes the number of hidden units.

RIRL gradients are different from BPTT gradients.