Ben Johnson

Yeah I bet that would do it... but have you ever seen a model that uses all of those techniques? I'm not sure _any_ datasets are properly benchmarked by your...

Interesting, thanks. Do you have a link to that paper, by chance?

On Omniglot, 5way/1shot and 20way/1shot they report 86.0% and 72.9% accuracies, respectively. I'm getting something like 95% and 90% from the baseline on those tasks. Note I'm not using the...

I'm talking about the baseline models, not the Matching Networks. ``` Model Matching Fn Fine Tune 5-way Acc 20-way Acc 1-shot 5-shot 1-shot 5-shot PIXELS Cosine N 41.7% 63.2% 26.7%...

2.89 uses cutout regularization, which `cifar10_micro_final.sh` does not use by default. Without cutout, the number in the paper is 3.54 -- but they don't measure the variance. When I trained...

Fantastic -- thank you. I'm running the `cifar10_micro_search.sh` now, and will post here to confirm once I get some results. ~ Ben

OK -- tail of `cifar10_micro_search.sh` looks like: ``` Eval at 42018 valid_accuracy: 0.6820 Eval at 42018 test_accuracy: 0.6636 epoch=149 ch_step=42050 loss=0.910298 lr=0.0005 |g|=2.4888 tr_acc=105/160 mins=717.36 epoch=149 ch_step=42100 loss=1.008317 lr=0.0005 |g|=3.0906...

Here's a plot of the test accuracy in `cifar10_micro_final.sh`:  This used architectures: ``` fixed_arc="1 3 1 1 1 4 0 4 1...

Is geolocation information visible somewhere in the UI? I'm looking at the requests / DOM and I don't see it, but that doesn't mean it's not there.

A V100 vs the 1080ti will make a big difference I believe, because of the half precision stuff