1. Addendum to Section 5.2, "Basics of Matrix Manipulation and Calculus":
   List of handy formulas for gradients and Jacobian
2. Addendum to Section 5.2, "Basics of Matrix Manipulation and Calculus":
   Formulas for matrix inversion in block form
3. Addendum to Section 5.5, "Recursive Least-Squares Estimator":
   Incremental formula for LSE error measures
4. Addendum to Section 5.5, "Recursive Least-Squares Estimator":
   Recursive LSE in the number of parameters
5. Addendum to Section 5.8, "LSE for Nonlinear Models":
   List of nonlinear models that can be transformed into linear ones
6. Addendum to Ch 10.9.2:
   Sec10.9.2.ps (PostScript file)
   This is Eiji Mizutani's paper at WCCI (World Congress on Computational Intelligence) 1998. The paper points out a connection between reinforce learning and recurrent NNs to solve non-Markovian problems. A simple concrete simulation example is described in the paper.
7. Addendum to "Policy-only Learning" in Chapter 10:
   Sample path-based policy-only learning by actor neural networks
8. The following two papers are supplementary materials on the trust-region methods on page 139 and Section 6.8:
   • Powell's dogleg trust-region steps with the quasi-Newton augmented Hessian for neural nonnlinear least-squares learning, IJCNN'99
   • Computing Powell's dogleg steps for solving adaptive networks nonlinear least-squares problems, IFSA'99

4,685 page hits since Aug 24, 1998