Closed plonerma closed 3 years ago
PR #14 contains a starting point
Conclusion via b377133951b653faf52e92ea3f4b631338002463
Making maml-tf2 public depends heavily on #11, maybe we should just close this one and move the maml-tf2 stuff together into #11?
I added the task to #11 One thing I would also like to add to the comparison, is an empirical comparison (ie. accuracy numbers on omniglot and mini-imagenet as reported in the papers). I will close this issue once Im done with that.
Before you leave: After letting your browser compute a bunch of meta-updates, you may want to cool it off and take a look at our computationally lightweight but not-to-miss further reading section.
Top-notch transition by the way! I like it :smile:
One thing I would also like to add to the comparison, is an empirical comparison (ie. accuracy numbers on omniglot and mini-imagenet as reported in the papers).
Do you want to add screenshots or build your own table?
I will just translate it into html. One method (MAML, FOMAML, REPTILE, iMAML) per row, and one or two columns per task. I will use the results from the papers though.
I think this might integrate nicer then a screenshot
Before we didn't include empirical results, which (in my opinion) is actually good, since we are focusing on the theory. But I think it would be nice to show at the end: the methods are different, but all produce reasonably good results...
=> That might also be a good justification of why we even cover all of them: there is no clear winner.
Empricial results added in 052fdcf35f7290b2207e268f7b8ca9f82b543827