Maximum likelihood training of probabilistic neural networks

A maximum likelihood method is presented for training probabilistic neural networks (PNN's) using a Gaussian kernel, or Parzen window. The proposed training algorithm enables general nonlinear discrimination and is a generalization of Fisher's method for linear discrimination. Important features of maximum likelihood training for PNN's are: 1) it economizes the well known Parzen window estimator while preserving feedforward NN architecture, 2) it utilizes class pooling to generalize classes represented by small training sets, 3) it gives smooth discriminant boundaries that often are "piece-wise flat" for statistical robustness, 4) it is very fast computationally compared to backpropagation, and 5) it is numerically stable. The effectiveness of the proposed maximum likelihood training algorithm is assessed using nonparametric statistical methods to define tolerance intervals on PNN classification performance.     

The complexity of on-line simulations between multidimensional turing machines and random access machines

To study different implementations of arrays, we present four results on the time complexities of on-line simulations between multidimensional Turing machines and random access machines (RAMs). First, everyd-dimensional Turing machine of time complexityt can be simulated by a log-cost RAM running inO(t(logt)^1–(1/d)(log logt)^1/d) time. Second, everyd-dimensional Turing machine of time complexityt can be simulated by a unit-cost RAM running inO(t/(logt)^1/d) time, provided that the input length iso(t/(logt)^1/d). Third, there is a log-cost RAMR of time complexityO(n), wheren is the input length, such that, for anyd-dimensional Turing machineM that simulatesR on-line,M requires (n ^{1 + (1/d)})/(logn(log logn)^{1 + (1/d)})) time. Fourth, every unit-cost RAM of time complexityt can be simulated by ad-dimensional Turing machine inO(t ²(logt)^1/2) time ifd = 2, and inO(t ²) time ifd 3. This result uses the weight-balanced trees of Nievergelt and Reingold.This paper was prepared while M. C. Loui was visiting the National Science Foundation in Washington, DC, and the Institute for Advanced Computer Studies, University of Maryland, College Park, MD. The views, opinions, and conclusions in this paper are those of the authors and should not be construed as an official position of the National Science Foundation, Department of Defense, U.S. Air Force, or any other U.S. government agency. The research of M. C. Loui was supported by the National Science Foundation under Grant CCR-8922008.     

Charge Recombination Dynamics in Organic Photovoltaic Systems with Enhanced Dielectric Constant

Increasing the dielectric constant of organic photovoltaic materials to reduce recombination rates has long been pursued, however, material modification often results in the modification of multiple device characteristics, making system comparison difficult. In this study, a fullerene derivative with an increased blend dielectric constant is examined by the addition of a triethylene glycol appendage to the fullerene (TEG‐PCBM). Density functional theory calculations show a small change to the permanent dipole moment between TEG‐PCBM and [6,6]‐phenyl‐C₆₁‐butyric acid methyl ester (PC₆₀BM) resulting in similar solubility, morphology, and device performance. TEG‐PCBM is blended with donors P3HT and PTB7‐Th and a comparable performance to PC₆₀BM is found. This model system shows the rarely reported characteristic of an increase in the dielectric constant while leaving its other properties unaltered. Looking at light intensity effects on open‐circuit voltage (V_oc), short‐circuit current (J_sc), and fill factor (FF) along with exciton dissociation efficiency, it is observed that when switching to the TEG‐ modified fullerene derivative, geminate recombination is not reduced, and Shockley–Read–Hall recombination is increased. While triethlyene glycol appendages may prove to be ineffective in improving recombination through increased dielectric constant, an approach for studying recombination in future high dielectric systems is provided.     

Solution‐Processed Semitransparent Organic Photovoltaics: From Molecular Design to Device Performance

Recent research efforts on solution‐processed semitransparent organic solar cells (OSCs) are presented. Essential properties of organic donor:acceptor bulk heterojunction blends and electrode materials, required for the combination of simultaneous high power conversion efficiency (PCE) and average visible transmittance of photovoltaic devices, are presented from the materials science and device engineering points of view. Aspects of optical perception, charge generation–recombination, and extraction processes relevant for semitransparent OSCs are also discussed in detail. Furthermore, the theoretical limits of PCE for fully transparent OSCs, compared to the performance of the best reported semitransparent OSCs, and options for further optimization are discussed.     

Estimating the parameters of general frequency modulated signals

A general frequency modulated (GFM) signal characterizes the vibrations produced by compressors, turbines, propellers, gears, and other rotating machines in a dynamic environment. A GFM signal is defined as the composition of a real or complex, periodic, or almost-periodic carrier function with a real, differentiable modulation function. A GFM signal therefore contains sinusoids whose frequencies are (possibly nonintegral) multiples of a fundamental; to distinguish a GFM signal from a set of unrelated sinusoids, it is necessary to track them as a group. This paper develops the general frequency modulation tracker (GFMT) for one or more GFM signals in noise using the expectation/conditional maximization (ECM) algorithm that is an extension of the expectation-maximization (EM) algorithm. Three advantages of this approach are that the ratios (harmonic numbers) of the carrier functions do not need to be known a priori, that the parameters of multiple signals are estimated simultaneously, and that the GFMT algorithm exploits knowledge of the noise spectrum so that a separate normalization procedure is not required. Several simulated examples are presented to illustrate the algorithm's performance.     

A signal/noise analysis of quasi-static MR elastography

In quasi-static magnetic resonance elastography, strain images of a tissue or material undergoing deformation are produced. In this paper, the signal/noise (S/N) ratio [SNR] of elastographic strain images, as measured by a phase-contrast technique, is analyzed. Experiments are conducted to illustrate how diffusion-mediated signal attenuation limits maximum strain SNR in small displacement cases, while the imaging point-spread function limits large displacement cases. A simple theoretical treatment agrees well with experiments and shows how an optimal displacement encoding moment can be predicted for a given experimental set of parameters to achieve a maximum strain SNR. A further experiment demonstrates how the limitation on strain SNR posed by the imaging point-spread function may potentially be overcome.