Statistics

English Version mxnet
1

https://d2l.ai/chapter_appendix-mathematics-for-deep-learning/statistics.html

1 Like
2

If a model suffers from high bias error, we often say it is underfitting or lack of generalization

If a model suffers from high variance, we often say it is overfitting and lack of flexibility

I think the characteristics of a high bias model and a high variance model are mixed up here. I believe that a high bias model lacks of flexibility while a high variance model lacks of generalization.

3

Hey @rmn0ff, great catch! Can you point out the link for the accurate reference?

4

image
image1327×368 86.3 KB

https://keras.rstudio.com/articles/tutorial_overfit_underfit.html
I agree that a high variance model lacks of generalization , if it means that the model doesn’t work in other patterns.
But I still feel confused about the meaning of flexibility, if it means that the model doesn’t work in these data, then @rmn0ff is right.

https://www.researchgate.net/publication/333505702_The_Theory_Behind_Overfitting_Cross_Validation_Regularization_Bagging_and_Boosting_Tutorial

Page 7:

image
image1103×674 95.5 KB


image
image1013×486 72.2 KB

image
image1096×739 82.6 KB

Now, I think @rmn0ff is right.

5

I notice that we don’t have the corresponding discussion of pytorch and tensorflow versions to last chapters.
And I tried to add some. Is it right? @astonzhang, @goldpiggy
https://discuss.d2l.ai/t/how-to-create-a-discussion-about-the-books-chapter/552/3

6

Hi @StevenJokes, we are working on that!

7

Just be quicker. Or can I help you? How to add these?