New answers tagged

1

No, there are no guarantees. SGD finds a local optimum but not a global optimum, and the solution it finds can be arbitrarily bad, if you have an unfriendly objective function. The only results I've seen that provide guarantees start from strong assumptions about the objective function (e.g., convexity).


2

In my opinion, this may be best approached as a sequence tagging problem, similar to part-of-speech tagging or named entity recognition. (So, this would be the seq2seq option, rather than regular classification). For example, think about it as trying to decide for each token, whether it is the start of a new verse or not. The advantage of the sequence ...


0

I am skeptical that machine learning is the right tool for this problem. I would look for a more direct solution, perhaps using peak detection or changepoint detection, or some other form of classical method for time-series analysis. If you do use machine learning, cross-entropy loss is not the right loss for your problem, and you will definitely need to ...


2

That's a tough one. I don't know what could cause that. It can be hard to know why we see the results we do, when working with neural nets. Often the best we can do is form several hypotheses, and then devise experiments to try to test those hypotheses. Some possible explanations that you could try to test: Perhaps it is noise / random chance Perhaps the ...


0

You say that your goal is to create a model and compute the likelihood of a test sample using that model. For that purposes, you do not need to find a way to interpret the hidden states, and you do not need to engineer the model so the hidden states have a meaning that you can understand. Instead, let the standard Baum-Welch algorithm compute a reasonable ...


Top 50 recent answers are included