 |
ML Newsnew | past | best | rss | submit | about |
| Understanding Deep Convolutional Networks (arxiv.org) | |
| 4 points by max_ on Feb 10, 2016 | hide | past | pdf | discuss on HN | |
|
In plain words: This lays out the math behind deep networks that pass data through layers of filters and nonlinear steps. It shows their stable patterns come from shrinking data across scales, untangling repeated symmetries, and separating sparse signals — how they classify and predict on high-dimensional data. AbstractDeep convolutional networks provide state of the art classifications and regressions results over many high-dimensional problems. We review their architecture, which scatters data with a cascade of linear filter weights and non-linearities. A mathematical framework is introduced to analyze their properties. Computations of invariants involve multiscale contractions, the linearization of hierarchical symmetries, and sparse separations. Applications are discussed. | |
| add comment on HN |