For many decades, techniques from filter design and system identification have been used to estimate physical parameters of dynamic systems for purposes of simulation and/or control. In the current era, such methods are being extended to employ neural networks. As a result, previously intractable inverse problems are being solved amazingly well by means of local minima in extremely high-dimensional nonlinear optimizations. This presentation will review some highlights of this research and summarize our own efforts in this area to date. Latest results and references can be found on the CCRMA JOS Home Page: https://ccrma.stanford.edu/∼jos/#synth-matching.
Smith et al. (Wed,) studied this question.