This code implements a Gauss-Newton optimization of objective functions that can be iteratively approximated by quadratics. This approach is particularly appropriate for least-squares inversions of moderately non-linear transforms. You will also find code for conjugate-gradient and line-search optimizations.
Get documentation of the algorithm here: [ ../../papers/inv/inv.html ] [ ../../papers/inv.pdf ] [ ../../papers/inv.ps.gz ]
Several papers describe ways to use this code: [ ../../papers/regularization.pdf ] [ ../../papers/regularization/ ] [ ../../papers/neural.pdf ] [ ../../papers/neural/ ] [ ../../papers/rmsinv.pdf ] [ ../../papers/rmsinv/ ]
See an older C++ version [ ../conjugate_gradients/ ]
See the java documentation in the documentation subdirectory [ documentation/ ] .
The current version of this code is now a part of the Mines Java Toolkit at https://github.com/MinesJTK/jtk in the edu.mines.jtk.opt package, with code in https://github.com/MinesJTK/jtk/tree/master/core/src/main/java/edu/mines/jtk/opt and documentation in https://github.com/MinesJTK/jtk/tree/master/docs/opt_package
An older public version is available from http://code.google.com/p/optimal/