Neural Networks for Combinatorial Optimization: a Review of More Than a Decade of Research
INFORMS Journal on Computing
Hopfield Network as Static Optimizer: Learning the Weights and Eliminating the Guesswork
Neural Processing Letters
| Hi-index | 0.00 |
This paper proposes an innovative enhancement of the classical Hopfield network algorithm (and potentially its stochastic derivatives) with an "adaptation mechanism" to guide the neural search process towards high-quality solutions for large-scale static optimization problems. Specifically, a novel methodology that employs gradient-descent in the error space to adapt weights and constraint weight parameters in order to guide the network dynamics towards solutions is formulated. In doing so, a creative algebraic approach to define error values for each neuron without knowing the desired output values for the same is adapted.