Trading variance reduction with unbiasedness: the regularized subspace information criterion for robust model selection in kernel regression

A well-known result by Stein (1956) shows that in particular situations, biased estimators can yield better parameter estimates than their generally preferred unbiased counterparts. This letter follows the same spirit, as we will stabilize the unbiased generalization error estimates by regularization and finally obtain more robust model selection criteria for learning. We trade a small bias against a larger variance reduction, which has the beneficial effect of being more precise on a single training set. We focus on the subspace information criterion (SIC), which is an unbiased estimator of the expected generalization error measured by the reproducing kernel Hilbert space norm. SIC can be applied to the kernel regression, and it was shown in earlier experiments that a small regularization of SIC has a stabilization effect. However, it remained open how to appropriately determine the degree of regularization in SIC. In this article, we derive an unbiased estimator of the expected squared error, between SIC and the expected generalization error and propose determining the degree of regularization of SIC such that the estimator of the expected squared error is minimized. Computer simulations with artificial and real data sets illustrate that the proposed method works effectively for improving the precision of SIC, especially in the high-noise-level cases. We furthermore compare the proposed method to the original SIC, the cross-validation, and an empirical Bayesian method in ridge parameter selection, with good results.     

Speech recognition using pulse-coupled neural networks with a radial basis function

We have developed a novel pulse-coupled neural network (PCNN) for speech recognition. One of the advantages of the PCNN is in its biologically based neural dynamic structure using feedback connections. To recall the memorized pattern, a radial basis function (RBF) is incorporated into the proposed PCNN. Simulation results show that the PCNN with a RBF can be useful for phoneme recognition. This work was presented in part at the 7th International Symposium on Artificial Life and Robotics, Oita, Japan, January 16–18, 2002     

Diagnosis of right ventricular infarction: experimental study through the use of body surface isopotential maps

This investigation was designed to diagnose right ventricular infarction, which is difficult to diagnose by the standard twelve-lead ECG, through the use of body surface isopotential maps which have significant diagnostic information. Right ventricular infarction was experimentally caused by ligation of the canine right coronary artery. Each dog had a series of maps recorded before and a week after experimentally-induced myocardial infarction. The common features of maps in right ventricular infarction are: 1. In the early stage of the ventricular depolarization the negative area occupies a comparatively large part of the right anterior chest surface, and in the middle stage, the larger part of the right anterior surface is also occupied by the negative area. 2. A minimum appears on the right anterior chest surface in the early stage. The delayed excitation resulting from intraventricular conduction disturbance caused by infarction, as verified by the epicardial isochronic map, is also well represented by the body surface isopotential map. In conclusion, through the use of body surface isopotential maps, it is much easier to diagnose right ventricular infarction and intraventricular conduction disturbance caused by infarction even in cases in which the standard twelve-lead ECG does not show the abnormalities clearly.     

Low temperature deposition of hexagonal BN films by chemical vapour deposition

Transparent hexagonal BN films were deposited onto copper substrates from the reactant gas BCl₃-NH₃-H₂ at temperatures in the range 250–700°C. The lowest deposition temperature of the films was about 250°C. The films deposited at temperatures below 450°C were unstable in moist atmosphere and devitrified; a 20%–30% decrease in weight was observed when these films were heated above 600°C in an argon atmosphere. In contrast, the films deposited at temperatures above 600°C were very stable, decreased in weight by 1%–2% on heating and were stable in air at temperatures below 750°C.     

Subspace information criterion for model selection

The problem of model selection is considerably important for acquiring higher levels of generalization capability in supervised learning. In this article, we propose a new criterion for model selection, the subspace information criterion (SIC), which is a generalization of Mallows's C(L). It is assumed that the learning target function belongs to a specified functional Hilbert space and the generalization error is defined as the Hilbert space squared norm of the difference between the learning result function and target function. SIC gives an unbiased estimate of the generalization error so defined. SIC assumes the availability of an unbiased estimate of the target function and the noise covariance matrix, which are generally unknown. A practical calculation method of SIC for least-mean-squares learning is provided under the assumption that the dimension of the Hilbert space is less than the number of training examples. Finally, computer simulations in two examples show that SIC works well even when the number of training examples is small.     

Mask design compensation for sloped sidewall structures fabricated by X-ray lithography

We have investigated and report in this paper the factors influencing the deformation caused by the dependence between the absorbed X-ray energy on the resist and the shape of the absorber on the X-ray mask. Based on the measurement of errors that occurred during the transferring process between the 2-D shape of mask pattern and the resulting wall of the fabricated 3-D structure, we have developed newly useful graphical data on the absorbed X-ray energy, dosage, and shape of a microstructure. As a result, it is being reported as a method for compensation for the deformed shape after the fabrication of a quadruplets-microneedle. We have considered a number of factors affecting the deformation and finally realized that the effect of a dose–depth nonlinear curve is the most possible cause. Without the compensation of the mask design, we could observe the deformed shapes of the sloped sidewall on the exposed structures. Polymethylmethacrylate microneedle structures fabricated by X-ray lithography with an additional plane-pattern to cross-section transfers technique are directly influenced by the absorber on the X-ray mask pattern. The sidewall of the microneedle was improved by changing the mask pattern from a double right-triangular pattern to a double semi-circular pattern, modeled by comparing the results from a mask-pattern and the actual structure.     

Analytical Model for Redistribution Profile of Ion-Implanted Impurities During Solid-Phase Epitaxy

We evaluated the redistribution profiles of ion-implanted impurities during solid-phase epitaxy using Rutherford backscattering spectrometry (RBS). RBS data revealed that the As concentration changes only near the moving amorphous/crystal interface. We derived an analytical model for the redistribution profiles using a segregation coefficient m between amorphous and crystalline Si, introduced a parameter of reaction length l that is the distance where impurities were exchanged, and obtained good agreement with experimental data with an m value of 3 and an l value of 1 nm for As. Furthermore, we applied our model to P and B redistribution profiles and obtained good agreement with corresponding m value of 4 and l value of 1 nm for P and m value of 0.3 and l value of 1 nm for B     

Characteristics of Linear Synchronous Motor Drive Cycloconverter for Maglev Vehicle ML-500 at Miyazaki Test Track

Maglev vehicle ML-500 in Japan attained a speed of 517 km/h on December 21, 1979. The linear synchronous motor (LSM) drive cycloconverter fabricated and submitted to the track test has a capacity of approximately 12 000 kVA in a continuously variable frequency range of 0-34 Hz with a sinusoidal current waveform. The current can be arbitrarily controlled in a range of 200-1300 A. Despite the rigorous power conversion and control, excellent current control characteristics were obtained: less than 4 percent in deviation of current peak value, about 4 ° in leading phase deviation and approximately 1.2 ms in zero current interval. Some observations about the LSM driving system are made, and an outline of the design and the data obtained from the super-high-speed running tests on Miyazaki test track are given.