Skip to content

Implementation and exploration of different optimization algorithms such as SGD, RMSprop, and ADAM.

Notifications You must be signed in to change notification settings

shunitavni/Optimization-algorithms

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 

Repository files navigation

Optimization-algorithms

Implementation and exploration of different optimization algorithms.

ADAM analysis

One of the complaints about ADAM is that still, the adjusted learning rates (see third math line in https://docs.google.com/presentation/d/1UFmncokDwUC4xLChHOfhe26MJucMyrjPjP9iLcKo30A/edit#slide=id.g5871948b3d_0_86) can be extream (very high or very low). Let's try to confirm this:

Modify your step_adam code to calculate the smallest and largest adjusted rates in the network in every step, and plot that as a curve (X axis is step number, and Y axis is the adjusted LR value). Plot two curves: one for min and one for max. What's your conclusions?

About

Implementation and exploration of different optimization algorithms such as SGD, RMSprop, and ADAM.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published