Neural Networks as Ordinary Differential Equations

Re:Neural Networks as Ordinary Differential Equations by itin on Monday December 17, 2018 @05:53PM (#18703006)

Is it just me or is the math here incorrect?

https://imgur.com/a/QXUZerV

[ Parent | on HN ]

Re:Neural Networks as Ordinary Differential Equations by itin on Monday December 17, 2018 @05:53PM (#18703012)

specifically, I don't see where the last step G(h_t) = etc. follows from

[ Parent | on HN ]

Re:Neural Networks as Ordinary Differential Equations by rkevingibson on Monday December 17, 2018 @05:55PM (#18703024)

Yup, that's a mistake. Shouldn't have the +h term at the end.

[ Parent | on HN ]

Re:Neural Networks as Ordinary Differential Equations by jeromebaek on Monday December 17, 2018 @06:27PM (#18703252)

Stared at this for a while also. Yes this is wrong.

[ Parent | on HN ]

Re:Neural Networks as Ordinary Differential Equations by _fullpint on Tuesday December 18, 2018 @12:21AM (#18704700)

Thank you for bringing this up. Legit, was bothering me to the extent I didn’t believe I could comprehend the article.

[ Parent | on HN ]

Re:Neural Networks as Ordinary Differential Equations by vanderZwan on Tuesday December 18, 2018 @06:30AM (#18705815)

So what is the correct equation? Or has the article already been fixed?

[ Parent | on HN ]

Re:Neural Networks as Ordinary Differential Equations by mlevental on Tuesday December 18, 2018 @11:00AM (#18707500)

the mistake is simply that G(h(t)) = F(h(t))/dt, rather than G(h(t)) = F(h(t))/dt + h(t), i.e. it's only in the text following the inset equations. you can verify this by looking at the "Evaluating ODEs" section.

edit: looks to have been fixed now.

[ Parent | on HN ]

Re:Neural Networks as Ordinary Differential Equations by sparkcc on Tuesday December 18, 2018 @06:19AM (#18705771)

Thanks a lot. reading

[ Parent | on HN ]