I do not Unterstand the concept of multiple units in lstm.
If i have an lstm layer with 64 cells, how would be the cells applied to each time step by unrolling.
My understanding is that each time step would be applied by unrolling to all cells.
So If unrolling equals 5, all five time steps would be applied in total to 5*64..
Is this correct?
all five time steps would be applied in total to 5*64..
Don't have rep to comment, but I'm not sure what you mean by "in total." If you are referring to the total amount of times a cell in the LSTM layer is updated, then you would be correct.
Yes, every time step, every cell in the LSTM layer is updated, but it is important to note that weights in multi-celled LSTM layers are passed differently than single-celled LSTM layers. Something that may be hard to understand is that at every time step, LSTM cells do not pass hidden weights to only their cell in the next time step. Diagrams like these can be misleading because they only show memory cells with hidden weights connecting to their own cells. In reality, a memory cell has a hidden weight for every cell in that layer. This makes the calculations for LSTM layers much easier since hidden weights can now be represented as matrices.
