Notes
- random linear combinations of the hidden units in the final layer of a neural network are semantically indistinguishable from the units themselves
Papers
- How transferable are features in deep neural networks? http://arxiv.org/pdf/1411.1792v1.pdf