Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
mjburgess
on Dec 6, 2022
|
parent
|
context
|
favorite
| on:
Turing Machines Are Recurrent Neural Networks (199...
Their derivatives, f^n(x) for arbitrary n, do not converge fine. Indeed, even n=1 is usually terrible in practice.
Fit a pleasant day-trip velocity curve with a NN and the acceleration would kill you.
whatshisface
on Dec 6, 2022
[–]
Wouldn't that be mostly a consequence of the loss function used to compute the distance between the curve and the model output?
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search:
Fit a pleasant day-trip velocity curve with a NN and the acceleration would kill you.