Questions

1. In an autoencoder, both the encoder and decoder must be structurally symmetric. Is this correct?
2. Given a dataset, X, and its transformation, Y, based on the code yielded by an autoencoder, all of the information contained in X can be found in Y. Is this correct?
3. A code, z_i ∈ (0, 1)¹²⁸ has sum(z_i) = 36. Is it sparse?
4. If std(z_i) = 0.03, is the code sparse?
5. A Sanger's network requires covariance matrix columns as input vectors. Is this correct?
6. How can we determine the importance of each component extracted by a Rubner-Tavan network?
7. Given a random vector, h_i ∈ ℜ^m (m is the output dimensionality of a DBN), is it possible to determine the most likely corresponding input sample?