New notes on using ML to generalize from small amounts of training data: attacking MNIST with just 10 training examples for each digit. 93.81% accuracy: http://cognitivemedium.com/rmnist_anneal_ensemble …
Show this thread
-
so then you are just averaging covnets with different hyperparameters?
No, with the same hyperparameters. (This is not uncommon.) The difference that arises is due to differences in initial weights. Using different convnets with different hps would certainly be worthwhile, but I haven't done it.
-
-
(The main focus, as you can probably tell from my post, was the annealing, not the use of the ensemble. I wanted to find the best possible hyper-parameters for a single net. )
-
oh ya totally. It just happened that last night I was thinking of how to ensemble covnets and cuoldnt quite get my hands around the idea since as ive said originally I always thought you only avrg very different models.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.