| |
Abstract:
Backpropagation networks process their inputs in a bottom-up
fashion, and use top-down connections to propagate an error signal
for learning. We introduce a new algorithm called up-propagation,
which uses top-down connections to generate patterns, and bottom-up
connections to propagate an error signal. The error signal is part
of a computational feedback loop that adjusts the generated pattern
to match sensory input. The error signal is also used for
unsupervised learning. The algorithm is benchmarked against
principal component analysis in experiments on images of
handwritten digits.
|
