Abstract
We show analytically that the expected number of fixed-point attractors in an associative memory neural network with analog neurons decreases exponentially as the neuron gain is reduced. Eliminating fixed-point attractors by using analog neurons has beneficial effects similar to stochastic annealing but can be easily implemented in a deterministic dynamical system such as an analog electronic circuit. Numerical data based on fixed-point counts in small networks support the analytical results.
- Received 20 November 1989
DOI:https://doi.org/10.1103/PhysRevLett.64.1986
©1990 American Physical Society