Shared Parameters Across Time

An RNN does not learn a separate model for each position.

The same update rule is reused at every time step. This sharing lets the model handle sequences of different lengths and keeps the parameter count independent of sequence length.

If the same cell is used for 10 time steps, the parameters are reused 10 times. The computation is repeated; the parameter set is shared.

Why this matters

Shared parameters make recurrence a sequence model rather than a fixed-position table. The model can learn a pattern such as "a modifier changes the next noun" and apply it at many positions.

Exercise

If one recurrent cell is applied across 6 time steps, how many distinct cell parameter sets are used?