Saul B. Gelfand

An iterative growing and pruning algorithm for classification tree design

A critical issue in classification tree design-obtaining right-sized trees, i.e. trees which neither underfit nor overfit the data-is addressed. Instead of stopping rules to halt partitioning, the approach of growing a large tree with pure terminal nodes and selectively pruning it back is used. A new efficient iterative method is proposed to grow and prune classification trees. This method divides the data sample into two subsets and iteratively grows a tree with one subset and prunes it with the other subset, successively interchanging the roles of the two subsets. The convergence and other properties of the algorithm are established. Theoretical and practical considerations suggest that the iterative free growing and pruning algorithm should perform better and require less computation than other widely used tree growing and pruning algorithms. Numerical results on a waveform recognition problem are presented to support this view. >

Classification trees with neural network feature extraction

The ideal use of small multilayer nets at the decision nodes of a binary classification tree to extract nonlinear features is proposed. The nets are trained and the tree is grown using a gradient-type learning algorithm in the multiclass case. The method improves on standard classification tree design methods in that it generally produces trees with lower error rates and fewer nodes. It also reduces the problems associated with training large unstructured nets and transfers the problem of selecting the size of the net to the simpler problem of finding a tree of the right size. An efficient tree pruning algorithm is proposed for this purpose. Trees constructed with the method and the CART method are compared on a waveform recognition problem and a handwritten character recognition problem. The approach demonstrates significant decrease in error rate and tree size. It also yields comparable error rates and shorter training times than a large multilayer net trained with backpropagation on the same problems.

Detection thresholds for tracking in clutter--A connection between estimation and signal processing

In the Kalman-Bucy filter and other trackers, the dependence of tracking performance upon the quality of the measurement data is well understood in terms of the measurement noise covariance matrix, which specifies the uncertainty in the values of the measurement inputs. The measurement noise and process noise covariances determine, via the Riccati equation, the state estimation error covariance. When the origin of the measurements is also uncertain, one has the widely studied problem of data association (or data correlation), and tracking performance depends critically on signal processing parameters, primarily the probabilities of detection and false alarm. In this paper we derive a modified Riccati equation that quantifies (approximately) the dependence of the state error covariance on these parameters. We also show how to use a receiver operating characteristic (ROC) curve in conjunction with the above relationship to determine the detection threshold in the signal processing system that provides measurements to the tracker so as to minimize tracking errors. The approach presented in this paper provides a feedback mechanism from the information processing (tracking) subsystem to the signal processing subsystem so as to optimize the overall performance in clutter.

Recursive stochastic algorithms for global optimization in R d

An algorithm of the form

A cost minimization approach to edge detection using simulated annealing

X_{k + 1} = X_k - a_k (\nabla U(X_k ) + \xi _k ) + b_k W_k

Reduced complexity decision feedback equalization for multipath channels with large delay spreads

, where

Simulated annealing with noisy or imprecise energy measurements

U( \cdot )

Metropolis-type annealing algorithms for global optimization in R d

is a smooth function on

Analysis of intertone and interblock interference in OFDM when the length of the cyclic prefix is shorter than the length of the impulse response of the channel

\mathbb{R}^d

Analysis of simulated annealing for optimization

,