4 ms·
> This is part of why deep learning and tensor methods work so well on image classification tasks. That statement sounds informative. Do you have a good non-s
by delhanty 11y ago
> This is part of why deep learning and tensor methods work so well on image classification tasks.
That statement sounds informative. Do you have a good non-specialist's reference?
- jamessb 11y agoYou might be interested by colah's (Christopher Olah's) blog: https://colah.github.io/ https://colah.github.io/ Two articles in particular are good introductions to looking at neural networks in terms of higher dimensional data lying on lower-dimesnional manifolds: Neural Networks, Manifolds, and Topology: https://colah.github.io/posts/2014-03-NN-Manifolds-Topology/ https://colah.github.io/posts/2014-03-NN-Manifolds-Topology/ Visualizing MNIST: An Exploration of Dimensionality Reduction: https://colah.github.io/posts/2014-10-Visualizing-MNIST/ https://colah.github.io/posts/2014-10-Visualizing-MNIST/
- delhanty 11y agoThank you for the useful links.