Search

Measurement noise is an integral part while collecting data of a physical process. Thus, noise removal is a necessary step to draw conclusions from these data, and it often becomes quite essential to construct dynamical models using these data. We discuss a methodology to learn differential equation(s) using noisy and sparsely sampled measurements. In our methodology, the main innovation can be seen in of integration of deep neural networks with a classical numerical integration method. Precisely, we aim at learning a neural network that implicitly represents the data and an additional neural network that models the vector fields of the dependent variables. We combine these two networks by enforcing the constraint that the data at the next time-steps can be given by following a numerical integration scheme such as the fourth-order Runge-Kutta scheme. The proposed framework to learn a model predicting the vector field is highly effective under noisy measurements. The approach can handle scenarios where dependent variables are not available at the same temporal grid. We demonstrate the effectiveness of the proposed method to learning models using data obtained from various differential equations. The proposed approach provides a promising methodology to learn dynamic models, where the first-principle understanding remains opaque.

In this paper, we propose an improved method for computing the $\mathcal{H}_\infty$ norm of linear dynamical systems that results in a code that is often several times faster than existing methods. Our approach uses standard optimization tools to rebalance the work load of the standard algorithm due to Boyd, Balakrishnan, Bruinsma, and Steinbuch, with the aim of minimizing the number of expensive eigenvalue computations that must be performed. Unlike the standard algorithm, our improved approach can also calculate the $\mathcal{H}_\infty$ norm to full precision with little extra work, and also offers some opportunity to improve its performance via parallelization. Finally, our improved method is also applicable for approximating the $\mathcal{H}_\infty$ norm of large-scale systems.

We evaluate different algorithms and the use of a mixed-precision approach for the solution of Algebraic Riccati Equations (AREs). The mixed-precision method obtains an approximation to the solution using single-precision arithmetic and then, this approximation is improved via a cheap iterative refinement. Some numerical results show that the mixed-precision solver reports time and energy savings and also provides similar or even more accurate solutions than well-known methods like the Sign Function or SDA on CPU-GPU platforms.

We discuss a Krylov-Schur like restarting technique applied within the symplectic Lanczos algorithm for the Hamiltonian eigenvalue problem. This allows to easily implement a purging and locking strategy in order to improve the convergence properties of the symplectic Lanczos algorithm. The Krylov-Schur-like restarting is based on the SR algorithm. Some ingredients of the latter need to be adapted to the structure of the symplectic Lanczos recursion. We demonstrate the efficiency of the new method for several Hamiltonian eigenproblems.

Delirium is a frequent disorder in the ICU associated with poor outcome. Several organizations, including the SCCM and the APA, therefore recommend standard screening for delirium to improve early diagnosis and treatment. The CAM-ICU is the most frequently used delirium detection tool in the Netherlands. Numerous studies to validate the CAM-ICU always yielded excellent sensitivity and specificity, but were all performed by a limited number of research nurses.

We consider the numerical solution of quadratic eigenproblems with spectra that exhibit Hamiltonian symmetry. We propose to solve such problems by applying a Krylov-Schur-type method based on the symplectic Lanczos process to a structured linearization of the quadratic matrix polynomial. In order to compute interior eigenvalues, we propose several shift-and-invert operators with Hamiltonian structure. Our approach is tested for several examples from structural analysis and gyroscopic systems.

SLICOT is a comprehensive numerical software package for control systems analysis and design. While based on highly performant Fortran routines, Matlab and Scilab interfaces provide convenient alternative access for users. In this survey, we summarize the functionality contained in the three SLICOT toolboxes for (1) basic tasks in systems and control, (2) system identification, and (3) model and controller reduction. Several examples illustrate the use of these toolboxes for addressing frequent computational tasks. © Oldenbourg Wissenschaftsverlag.