Simulating dynamic covariance structures for testing the adaptive behavior of variable selection algorithms (Invited Paper)

Variable selection for regression is a classical statistical problem, motivated by concerns that too large a number of covariates may bring about overfitting and unnecessarily high measurement costs. Novel difficulties arise in streaming contexts, where the correlation structure of the process may be drifting, in which case it must be constantly tracked so that selections may be revised accordingly. A particularly interesting phenomenon is that non-selected covariates become missing variables, inducing bias on subsequent decisions. This raises an intricate exploration-exploitation tradeoff, whose dependence on the covariance tracking algorithm and the choice of variable selection scheme is too complex to be dealt with analytically. We hence capitalise on the strength of simulations to explore this problem, taking the opportunity to tackle the difficult task of simulating dynamic correlation structures. © 2008 IEEE.

Fine scale structures from deformation of aluminium containing small alumina particles

Microstructures and mechanical properties have been studied in aluminium containing a fine dispersion of alumina particles, deformed by cold-rolling to strains between 1.4 and 3.5. The microstructure was characterised by TEM. The deformation structures evolved very rapidly, forming a nanostructured material, with fine subgrains about 0.2 μm in diameter and a fraction of high-angle boundaries which was already high at a strain of 1.4, but continued to increase with rolling strain. The yield stress and ductility of the rolled materials were measured in tension, and properties were similar for all materials. Yield stress measurements were correlated with estimates made using microstructural models. The role of small particles in forming and stabilising the deformation structure is discussed. This nanostructured cold-deformed alloy has mechanical properties which are usefully enhanced at comparatively low cost. This gives it, and similar particle-strengthened alloys, good potential for commercial exploitation. © 2002 Acta Materialia Inc. Published by Elsevier Science Ltd. All rights reserved.

Efficiency measurement of GaN-based quantum well and light-emitting diode structures grown on silicon substrates

The optical efficiency of GaN-based multiple quantum well (MQW) and light emitting diode (LED) structures grown on Si(111) substrates by metal-organic vapor phase epitaxy was measured and compared with equivalent structures on sapphire. The crystalline quality of the LED structures was comprehensively characterized using x-ray diffraction, atomic force microscopy, and plan-view transmission electron microscopy. A room temperature photoluminescence (PL) internal quantum efficiency (IQE) as high as 58% has been achieved in an InGaN/GaN MQW on Si, emitting at 460 nm. This is the highest reported PL-IQE of a c-plane GaN-based MQW on Si, and the radiative efficiency of this sample compares well with similar structures grown on sapphire. Processed LED devices on Si also show good electroluminescence (EL) performance, including a forward bias voltage of ∼3.5 V at 20 mA and a light output power of 1 mW at 45 mA from a 500 ×500 μm2 planar device without the use of any additional techniques to enhance the output coupling. The extraction efficiency of the LED devices was calculated, and the EL-IQE was then estimated to have a maximum value of 33% at a current density of 4 A cm-2, dropping to 30% at a current density of 40 A cm-2 for a planar LED device on Si emitting at 455 nm. The EL-IQE was clearly observed to increase as the structural quality of the material increased for devices on both sapphire and Si substrates. © 2011 American Institute of Physics.

Normally closed microgrippers based on diamond-like carbon/Ni bimorph and diamond-like carbon/metal/polymer trimorph structures

Multi-finger, normally-closed microgrippers made from a bilayer of a metal and diamond-like carbon (DLC) or a trilayer of a polymer, metal and DLC have been analysed, simulated and fabricated. Temperatures of ∼700 K are necessary to open Ni/DLC bimorph structures. Microgrippers made from an SU8/DLC bilayer or SU8/Al/DLC trilayer have also been fabricated, and fully closed microcages with diameters of ∑40 μm have been obtained. Using SU8 reduces the opening temperature of these devices to only ∼400 K.

Three-dimensional carbon nanowall structures

The authors report the growth of carbon nanowalls in freestanding, three-dimensional aggregates by microwave plasma-enhanced chemical vapor deposition. Carbon nanowalls extrude from plasma sites into three-dimensional space. The growth is catalyst-free and not limited by nucleating surfaces. The growth mechanism is discussed and compared with similar carbon nanomaterials. High surface area of as-grown carbon nanowalls indicates a potential for electrochemical applications. Field emission measurements show a low field turn-on and long-term stability. The results establish a scalable production method and possible applications using field emission or high surface area. © 2007 American Institute of Physics.

Flux closure vortexlike domain structures in ferroelectric thin films

Enhanced piezoresponse force microscopy was used to study flux closure vortexlike structures of 90° ferroelastic domains at the nanoscale in thin ferroelectric lead zirconium titanate (PZT) films. Using an external electric field, a vortexlike structure was induced far away from a grain boundary, indicating that physical edges are not necessary for nucleation contrary to previous suggestions. We demonstrate two different configurations of vortexlike structures, one of which has not been observed before. The stability of these structures is found to be size dependent, supporting previous predictions. © 2010 The American Physical Society.

Self-aligned electrodes for suspended carbon nanotube structures

We report on a study into electrode fabrication for the gate control of carbon nanotubes partially suspended above an oxidised silicon substrate. A fabrication technique has been developed that allows self-aligned side-gate electrodes to be placed with respect to an individual nanotube with a spacing of less than 10 nm. The suspended multi-walled carbon nanotube (MWCNT) is used as an evaporation mask during metal deposition. The metal forms an island on the nanotube, with increasing width as the metal is deposited, forming a wedge shape, so that even thick deposited layers yield islands that remain separated from the metal deposited on the substrate due to shadowing of the evaporation. The island can be removed during lift-off to leave a set of self-aligned electrodes on the substrate. Results show that Cr yields self-aligned side gates with around 90% effectiveness. © 2003 Elsevier Science B.V. All rights reserved.

Study of multi-walled carbon nanotube structures fabricated by PMMA suspended dispersion

We report on the electrical characteristics of plasma enhanced chemical vapour deposition (PECVD)-grown, multi-walled carbon nanotube (MWCNT) devices made by a new fabrication method, PMMA suspended dispersion. This method makes it possible to suspend nanotubes between metal electrodes and to remove unwanted nanotubes from the substrate. The measurements show that the MWCNTs are metallic and able to maintain a current density ∼2×106 A/cm2 for more than 15 days with a maximum current density of ∼1.8×107 A/cm2. This high current density and reliability will make PECVD-grown MWCNTs applicable to field emission cathodes. © 2002 Elsevier Science B.V. All rights reserved.

Tractable nonparametric bayesian inference in poisson processes with Gaussian process intensities

The inhomogeneous Poisson process is a point process that has varying intensity across its domain (usually time or space). For nonparametric Bayesian modeling, the Gaussian process is a useful way to place a prior distribution on this intensity. The combination of a Poisson process and GP is known as a Gaussian Cox process, or doubly-stochastic Poisson process. Likelihood-based inference in these models requires an intractable integral over an infinite-dimensional random function. In this paper we present the first approach to Gaussian Cox processes in which it is possible to perform inference without introducing approximations or finitedimensional proxy distributions. We call our method the Sigmoidal Gaussian Cox Process, which uses a generative model for Poisson data to enable tractable inference via Markov chain Monte Carlo. We compare our methods to competing methods on synthetic data and apply it to several real-world data sets. Copyright 2009.

Tractable nonparametric Bayesian inference in Poisson processes with Gaussian process intensities

The inhomogeneous Poisson process is a point process that has varying intensity across its domain (usually time or space). For nonparametric Bayesian modeling, the Gaussian process is a useful way to place a prior distribution on this intensity. The combination of a Poisson process and GP is known as a Gaussian Cox process, or doubly-stochastic Poisson process. Likelihood-based inference in these models requires an intractable integral over an infinite-dimensional random function. In this paper we present the first approach to Gaussian Cox processes in which it is possible to perform inference without introducing approximations or finite-dimensional proxy distributions. We call our method the Sigmoidal Gaussian Cox Process, which uses a generative model for Poisson data to enable tractable inference via Markov chain Monte Carlo. We compare our methods to competing methods on synthetic data and apply it to several real-world data sets.