Vote counting measures for ensemble classifiers

Various measures, such as Margin and Bias/Variance, have been proposed with the aim of gaining a better understanding of why Multiple Classifier Systems (MCS) perform as well as they do. While these measures provide different perspectives for MCS analysis, it is not clear how to use them for MCS design. In this paper a different measure based on a spectral representation is proposed for two-class problems. It incorporates terms representing positive and negative correlation of pairs of training patterns with respect to class labels. Experiments employing MLP base classifiers, in which parameters are fixed but systematically varied, demonstrate the sensitivity of the proposed measure to base classifier complexity.     

Convolution profiles for right inversion of multivariable non-minimum phase discrete-time systems

The problem of the non-causal inversion of linear multivariable discrete-time systems is analyzed in the geometric approach framework and is solved through the computation of convolution profiles which guarantee perfect tracking under the assumption of infinite-length preaction and postaction time intervals. It is shown how the shape of the convolution profiles is related to both the relative degree and the invariant zeros of the plant. A computational setting for the convolution profiles is derived by means of the standard geometric approach tools. Feasibility constraints are also taken into account. A possible implementation scheme, based on a finite impulse response system acting on a stabilized control loop, is provided.     

Fault-accommodation with intelligent sensors

Optimal sensor fault-accommodation is considered for faults modelled by an increase in measurement noise. This noise is taken to be bounded and its probabilistic properties unknown. It is assumed that intelligent (e.g. self-validating) instrumentation is in use and estimates of the noise bounds are available. The fault-tolerant controller is designed to optimize a noise rejection and a nominal reference tracking index and leads to a mixed norm minimization problem (l₁/l₂). We exploit known results and a particular feedback configuration to show that it is possible to optimize simultaneously without a trade-off the two performance indices. The results are applied to systems where the presence of auxiliary measurements allows for an optimal fault-accommodation strategy. Using properties of the optimal solution, we define a factorization for the optimal controller alternative to the Youla parametrization, leading to an algorithm which is optimal, transparent and efficient.     

Tracking trajectories of the cart-pendulum system

The problem of tracking a periodic trajectory of the well-known cart-pendulum system is solved. After a change of coordinates and a change of feedback, the equations of this system are nonlinear but feedforward. This property is extensively used to carry out for this system the design of uniformly asymptotically stabilizing time-varying state feedbacks by using the forwarding approach.     

A time aggregation approach to Markov decision processes

We propose a time aggregation approach for the solution of infinite horizon average cost Markov decision processes via policy iteration. In this approach, policy update is only carried out when the process visits a subset of the state space. As in state aggregation, this approach leads to a reduced state space, which may lead to a substantial reduction in computational and storage requirements, especially for problems with certain structural properties. However, in contrast to state aggregation, which generally results in an approximate model due to the loss of Markov property, time aggregation suffers no loss of accuracy, because the Markov property is preserved. Single sample path-based estimation algorithms are developed that allow the time aggregation approach to be implemented on-line for practical systems. Some numerical and simulation examples are presented to illustrate the ideas and potential computational savings.     

Algorithm for finding all k nearest neighbors

A very simple algorithm for computing all k nearest neighbors in 2-D is presented. The method does not rely on complicated forms of tessellation, it only requires simple data binning for fast range searching. Its applications range from scattered data interpolation to reverse engineering.