Which statements about LSTM are true?

a$.$ It is often beneficial to initialize the forget gate with positive bias values.

b$.$ The focused ISTM network feeds the input vector $x(t)$ only into input gate $i(t)$ and output gate $o(t)$ but not into the block input $z(t).$

c$.$ "Ticker steps" refers to the technique of letting an LSTM run for more time steps even after the input sequence has been taken in by the network.

d$.$ The inputs to the lightweight LSTM have to fulfill the Markov property.

e$.$ The forget gate reintroduces the vanishing gradient problem to a certain degree.

f$.$ To counter the drift effect, the input gate should be initialized with positive values.

$9.$ An LSTM with coupled input gate and forget gate, i$.$e$.,c(t)=(1-i(t))o.c(t-1)+i(t)o.z(t),$ leads $($alongside other changes$)$ to the gated recurrent unit $($GRU$).$

h$.$ Most LSTM networks used in practice do not feature a forget gate.