Numerical Simulation of Neural Networks with Translation and Rotation Invariant Pattern Recognition
Abstract
A Hopfield-like network modified for translation and rotation invariant pattern recognition is studied on a computer. The neural firing thresholds θ are used as additional degrees of freedom. The phase diagram in the space of parameters (T, θ) is obtained both for the translation and the rotation.
- Publication:
-
International Journal of Modern Physics B
- Pub Date:
- 1994
- DOI:
- 10.1142/S0217979294000658
- Bibcode:
- 1994IJMPB...8.1529A