I was just wondering why there is need of sampling, when a back propagation step adjusts the weights to predict a based on the loss function, which will predict a new character with the updated weights in a new example.
Which assignment in Week 1 are you asking about?
And what is your basis for linking backpropagation and sampling?
Just a small confusion in understanding the concept of sampling.
I thought, it is used during training time after SoftMax.
So sampling is used during Inference time to choose the best output among many generations based on several sampling methods( be it beam search or random choice etc) right?
Please do correct me if I am wrong or if there is a better way to frame it.
Thanks in advance.
I agree with your latest summary.