Is there a video lecture talking about how to find global minimum?

I have completed Machine Learning Specialization by the end of August. Today when I think about the content, I am wondering is there a video lecture talking about how to find global minimum while there are lots of local minimum? ( I think I have seen something in one of the video lectures, but now I couldn’t find it.) If you know which lecture it is, could you help point it out? :grinning: I want to watch it again.

No, there isn’t. Generally this isn’t a problem, because cost functions that have multiple minimums are usually very close together, so any one of them will usually work fine.

Keep in mind that we don’t necessarily always need the exact minimum cost. We just need a solution that gives good enough performance on a test set.

To check for certain, use the test sets results.

If you don’t get good enough results, the usual technique is to train again, using different initial random weights, and see if you get a lower cost. Repeat this process until you get a result that works well enough.

If you never get a good enough result, then go back to the beginning and use a more complex model.

2 Likes

Thank you so much!!!