No initialization of a0 in input to Bi-LSTM, week 3, ex 1

Course Q&A Deep Learning Specialization Sequence Models
1

Hi DeepLearning Team:

I noticed that in previous lessons we made sure to initialize a0, as we do with c0 and s0 in this exercise. However in this assignment we create a bidirectional LSTM with one line and I can’s see how we initialize a<0>, and a<Tx+1> to zero, even though we discuss that in the notes and it is illustrated that way in the figure, “Figure 1: Neural machine translation with attention”.

Is the input set to zero by default if no vector is provided?

Thanks, David

2

When using LSTM, 0s are assumed when no value is provided for initial_state

3

Thanks Balaji.
That’s good to know!
Best,
David