CSC321 Lecture 24

Using Boltzmann machines to initialize backpropagation

Some problems with backpropagation

A solution to all of these problems

Modelling the distribution of digit images

Results on permutation-invariant MNIST task

Why does greedy learning work?

Deep Autoencoders

The deep autoencoder

A comparison of methods for compressing digit images to 30 real numbers.

A very deep autoencoder for synthetic curves that only have 6 degrees of freedom

An autoencoder for patches of real faces

Reconstructions of face patches from new people

64 of the hidden units in the first hidden layer

How to find documents that are similar to a query document

How to compress the count vector

The non-linearity used for reconstructing bags of words

Performance of the autoencoder at document retrieval

Proportion of retrieved documents in same class as query

Slide 19

Slide 20

THE  END