4 ms·
I'd be interested in seeing if this same problem appears in generative models. It seems like a flaw of discriminative models. Convolutional Boltzmann machines
by jostmey 12y ago
I'd be interested in seeing if this same problem appears in generative models. It seems like a flaw of discriminative models.
Convolutional Boltzmann machines have been developed, and neural networks trained with Backpropagation can be pre-trained with generative models. Would this help alleviate the problem?
- Houshalter 12y agoThe original paper found that nets pretrained as autoencoders were the least susceptible to it, but still failed. IIRC the follow up paper (http://arxiv.org/abs/1412.6572 http://arxiv.org/abs/1412.6572) tested it or at least claimed it wasn't a solution.