Dynamical stability and parameter selection in neural optimization
Behzad Kamgar-Parsi, Behrooz Kamgar-Parsi · 1992
Finding suitable parameter values in solving optimization problems with the Hopfield net is of crucial importance. The authors present a systematic approach, based on analyzing dynamical stability of valid solutions for finding relationships among the parameters which make their selection much easier. This technique can show whether the problem formulation is flawed. As an example they discuss the Hopfield-Tank formulation of the traveling salesman problem and show that it is dynamically unstable, hence obtaining valid solutions is difficult. The modified formulation given by S.Y.B. Aiyer et al. (1990) which overcomes the instability problems is also discussed.>