PHD Discussions Logo

Ask, Learn and Accelerate in your PhD Research

Question Icon Post Your Answer

Question Icon

How can you prove the Mean Squared Error (MSE) surface is multimodal for a nonlinear system parameter estimation?

Why does the error surface in nonlinear parameter estimation often have many hills and valleys (i.e., is multimodal)?

All Answers (1 Answers In All)

By Jasmin Answered 8 months ago

To demonstrate that the Mean Squared Error (MSE) surface is multimodal for a nonlinear-in-parameters model, which is inherently non-convex, you can employ both analytical and empirical methods. Analytically, prove the existence of multiple stationary points where the gradient vanishes, including distinct local minima. Empirically, run the estimation algorithm (like gradient descent) from numerous random starting points; convergence to different parameter sets with comparable low MSE values strongly indicates multiple local minima, a result often corroborated by visualizing the cost surface for low-dimensional parameter subsets.

Your Answer