next up previous
Next: Acknowledgments Up: The Neural Heat Exchanger Previous: The Neural Heat Exchanger


The Neural Heat Exchanger is an entirely local method for training multi-layer nets. It is inspired by principles of conventional, physical heat exchangers. It was first presented in 1990, and is very similar to the 1995 Helmholtz Machine. It can solve certain non-linear tasks. To fully understand its advantages and limitations, however, much theoretical and empirical work remains to be done.

Juergen Schmidhuber 2003-02-28