Neural networks and genetic algorithms in the
solution of inverse problems

Keith A. Woodbury
Mechanical Engineering Department
The University of Alabama
Tuscaloosa, AL 35487

Both Neural Networks (NNs) and Genetic Algorithms (GAs) have been proposed and explored as tools to solve inverse problems. True inverse problems are those relying on imprecise data (generally collected via appropriate experimentation), and this imprecision reveals the ill-posedness of the inverse problem. In contrast, design problems rely on "perfect data" (the desired design condition) and thus are not ill-posed.

This paper summarizes the efforts of two prior investigations (Krejsa, et al. 1998, and Raudensky, et al., 1995) which employed NNs and GAs for in the inverse heat conduction problem. Other applications of these technologies to inverse problems are also mentioned. The conclusion is that use of these novel techniques does not alter the underlying ill-posed nature of the problem, and solution difficulties persist even when using non-gradient-based methods.