How does the neural network set the number of hidden layers and nodes? the more it is, the more accurate it will be.

I have seen several empirical formulas about the number of nodes in the hidden layer
94_200_200.jpg

1, for example, I am 40 input nodes and one output node. According to the empirical formula, if the hidden layer is single-layer hidden layer, the number of hidden layer nodes will not exceed 40, so isn"t it true that the more nodes are, the more accurate the number of nodes is? Or will it achieve the best results at some worthwhile time?
2. Similarly, if I have 40 inputs and one output, will two hidden layers iterate faster than one hidden layer? Why?
3. To determine whether the number of hidden layers and nodes in a single layer can only be tested step by step, is there any way to initialize these parameters quickly and fine-tune them at the beginning?
4. Theoretically, is it true that the more iterations, the more accurate the training model?

those who have an understanding of any of the above questions are welcome to answer, thank you very much


of course, the more layers you have, the longer the training time


, the more nodes, the better the effect on the training model, and even achieve 100% prediction accuracy. However, it brings the over-fitting of the model, and the prediction effect of putting the model on the test data is seriously reduced. Generally speaking, the effective depth of training should be considered so that the model can achieve a better prediction effect without over-fitting.


not necessarily the more, the better. I suggest you take a look at this. (I don't know if I just started watching machine learning. We all learn from each other.)

< machine learning > < < machine learning >

< machine learning. This answer is more professional, referring to the relationship between the number of layers and the number of nodes.
how the neural network selects the number of hidden layers

Menu