Second year
research advances
Research
Advances
Perturbed
dumped Newton and SQP-IP methods for
discretized optimal control problems
Researchers.
V. Ruggiero, S. Bonettini, F. Tinti.
Advances.
The convergence theory of the Interior Point (IP) method for NLP
problems as an Inexact Newton scheme
[BGR-05,
B-05]
has been developed; a
numerical comparison of different iterative solvers of the inner
linear system (Hestenes [BGR-04],
CG with indefinite preconditioners
[BR-05])
has been carried out.
Comparison between IP methods and
semismooth schemes for the numerical solution of variational
inequalities and realization of the related codes
[T-05].
[Top]
Variable
projection methods: a successful approach to
SVMs and large-scale simply constrained QP
Researchers.
L. Zanni, G. Zanghirati.
Advances.
For QP problems, the projected gradient methods have been employed as
inner solvers in decomposition techniques for training SVMs
[Z-06] using new line-search and steplength selection strategies.
In this framework, a new strategy for the selection of the working
set has been developed [SZZ-05a].
[Top]
Adaptive
diagonal space-filling curves: linking
geometric Lipschitz and Bayesian
approaches for non-differentiable global
optimization
Researchers.
Ya. D. Sergeyev, D. Lera, D. Kvasov.
Advances.
For the Global Optimization context, the new ideas are:
- construction of multidimensional support function
[MCGST-04]
for Interval Analysis methods
- employing of exponential sequences
[S-04a]
- feasible region partition strategy similar
to the one of the diagonal methods
[S-05],
with only one evaluation
of the objective function for each subinterval
- classification based on the computational complexity of minimization methods
[GL-05]
- decomposition algorithms, based on the uniform norm, for
Lipschitzian functions [GL-07]
- parallel methods for the global optimization
[Top]
Symbolic
calculus and parallel computing for error
propagation automatic control
Researchers.
G. Spaletta.
Advances.
In the Mathematica framework, the automatic
derivation of Runge-Kutta methods for the solution of ODEs has been
implemented [SS-04].
In [SS-06a,
SS-05c]
some methods exploiting the splitting
and the composition for the geometric integration of ODEs are
presented.
[Top]
Parallel
domain decomposition techniques for
space-variant blurred images restoration
Researchers.
F. Zama, E. Loli Piccolomini, G. Landi, M.
Bertaja.
Advances.
A descent method for the computation of the
solution and of the Tikhonov regularization parameter have been
applied to inverse problem coming from medical imaging, particularly
from SPECT and Dynamic Magnetic Resonance (DMR) data
[ZL-05,
ZLL-04,
LLZ-05] .
[Top]
A
parallel computing approach to dynamic
magnetic resonance imaging
Researchers.
F. Zama, E. Loli Piccolomini, G. Landi, M.
Bertaja.
Advances.
Some nonlinear regularization methods based on the
Total Variation principle have been
studied in
[LL-05,
LLZ-04] .
for the DMR problem.
|
