Re:AI Models Predict Breast Cancer with Radiologist-Level Ac...
by krisoft on Tuesday June 25, 2019 @05:18PM
(#20279061)
They haven't got erased, but more like subsumed? If you use dropout to train your model that is basically equivalent with using an ensemble of deep neural networks.
[ Parent
| on HN ]
Re:AI Models Predict Breast Cancer with Radiologist-Level Ac...
by irq11 on Wednesday June 26, 2019 @12:13AM
(#20281844)
That is not even close to the same thing.
If you train an ensemble of models with random dropout, you have an ensemble. Models trained with dropout will still have significant variation from run to run.
Re:AI Models Predict Breast Cancer with Radiologist-Level Ac...
by sooheon on Wednesday June 26, 2019 @12:24AM
(#20281903)
> That is not even close to the same thing.
It's a common interpretation: https://arxiv.org/abs/1706.06859
Re:AI Models Predict Breast Cancer with Radiologist-Level Ac...
by irq11 on Wednesday June 26, 2019 @06:53PM
(#20290099)
There may be a paper on it, but itβs not a common view.
In particular, this paper neglected to do the obvious thing: ensemble networks trained with dropout. It improves performance over dropout alone.