0
$\begingroup$

I was reading a paper about Neural Holography (page 5, equation 4), where authors used simple stochastic gradient descent as optimizing method. There I have encountered following update rule: enter image description here

, where alpha is a learning rate, L is a loss function. So basically they update parameter phi using both gradient of loss and loss itself. I have never encountered such an update rule. What is this update rule? They never specified, why they did it this way.

Improve this question
$\endgroup$
1
  • $\begingroup$ it's just an odd notation. They are doing regular gradient descent $\endgroup$
    – seanv507
    Commented Dec 29, 2023 at 17:30

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.