Seminar 5: Energy propagation in deep convolution neural networks

This is the 5th seminar in the weekly series of machine learning reading group (MLRG) at KTH. This presentation discusses the recent work on energy propagation and decay in deep convolutional neural networks, as means of providing fundamental limits on number of layers needed for feature extraction. The presentation is based on the following paper:

- Wiatowski, Thomas, Philipp Grohs, and Helmut Bölcskei. "Energy propagation in deep convolutional neural networks." arXiv preprint arXiv:1704.03636 (2017).
- Thomas Wiatowski, Helmut Bölcskei, "A Mathematical Theory of Deep Convolutional Neural Networks for Feature Extraction" (2015)

[jonabirdd]

Terrible video quality. Can barely look at the slides.