About this document

Understanding Deep Learning Insights by Edo Demirbilek is a document available to read on EtoBox.

1) Deep neural networks have been shown to learn increasingly complex features as more layers are added, with the last layer acting as a useful representation for other tasks. 2) While deep learning optimization is non-convex, stochastic gradient descent is able to find solutions that generalize well in practice, likely due to the "nice" geometry of the loss surface. 3) Increasing network size through depth, width, and regularization tricks improves both training and generalization performance, even as net

Author

Edo Demirbilek

Language

EN