Variable Sparsity Kernel Learning

This paper(1) presents novel algorithms and applications for a particular class of mixed-norm regularization based Multiple Kernel Learning (MKL) formulations. The formulations assume that the given kernels are grouped and employ l(1) norm regularization for promoting sparsity within RKHS norms of each group and l(s), s >= 2 norm regularization for promoting non-sparse combinations across groups. Various sparsity levels in combining the kernels can be achieved by varying the grouping of kernels-hence we name the formulations as Variable Sparsity Kernel Learning (VSKL) formulations. While previous attempts have a non-convex formulation, here we present a convex formulation which admits efficient Mirror-Descent (MD) based solving techniques. The proposed MD based algorithm optimizes over product of simplices and has a computational complexity of O (m(2)n(tot) log n(max)/epsilon(2)) where m is no. training data points, n(max), n(tot) are the maximum no. kernels in any group, total no. kernels respectively and epsilon is the error in approximating the objective. A detailed proof of convergence of the algorithm is also presented. Experimental results show that the VSKL formulations are well-suited for multi-modal learning tasks like object categorization. Results also show that the MD based algorithm outperforms state-of-the-art MKL solvers in terms of computational efficiency.

Novel photo-crosslinkable flame retardant polyvanillylidene arylphosphate esters

A new class of photo-crosslinkable flame retardant arylphosphate ester polymers based on diarylidenecycloalkanone groups has been synthesized by polymerizing 2,5-divanillylidene cyclopentanone and 2,6-divanillylidenecyclohexanone with various arylphosphorodichloridates by interfacial polycondensation using a phase transfer catalyst. The resulting polymers were characterized by inherent viscosity, g.p.c., i.r., H-1, C-13, P-31 n.m.r. spectroscopy. These polymers were studied for their photochemical and flame retardant properties. The divanillylidene cycloalkanone group in the chain function as photoactive centres while arylphosphate ester groups impart flame retardancy. The photo-crosslinking proceeds via 2 pi + 2 pi cycloaddition reaction of the divanillylidene cycloalkanone moieties. The crosslinking rate, thermal stability and flammability characteristics of the polymers increase with decrease in the size of the cycloalkanone ring. (C) 1997 Elsevier Science Ltd.

Polyethylene stibinite phosphate esters: Novel flame-retardant plasticizers for PVC

A new series of multielement flame-retardant plasticizers containing polyethylene stibinite phosphate esters have been prepared by bulk polymerization from ethylene glycol with various antimony (III) aryloxydichlorides and arylphosphorodichloridates possessing various combinations of substituent [Cl,Br,NO2]. All the polymers are pink-coloured viscous fluids. They were characterized by inherent viscosity, density, IR, H-1, C-13 and P-31 NMR spectroscopy. The thermal behaviour of the polymers was compared by thermogravimetric analysis and correlated with their structures. The flammability studies were carried out by the limiting oxygen index test. The polymers containing P, Sb, N and Pr elements in their backbone show superior thermal-and flame-retardant characteristics than the other polymers. A comparative study was carried out with one of the synthesized polymers as a polymeric flame-retardant additive to plasticized PVC. The results showed improved LOI and mechanical properties to that of the conventional flame-retardant additive composition. (C) 1997 Elsevier Science Ltd.

Robust Formulations for Handling Uncertainty in Kernel Matrices

We study the problem of uncertainty in the entries of the Kernel matrix, arising in SVM formulation. Using Chance Constraint Programming and a novel large deviation inequality we derive a formulation which is robust to such noise. The resulting formulation applies when the noise is Gaussian, or has finite support. The formulation in general is non-convex, but in several cases of interest it reduces to a convex program. The problem of uncertainty in kernel matrix is motivated from the real world problem of classifying proteins when the structures are provided with some uncertainty. The formulation derived here naturally incorporates such uncertainty in a principled manner leading to significant improvements over the state of the art. 1.

Large eddy simulation of a premixed flame with approximate deconvolution modeling

Approximate deconvolution modeling is a very recent approach to large eddy simulation of turbulent flows. It has been applied to compressible flows with success. Here, a premixed flame which forms in the wake of a flameholder has been selected to examine the subgrid-scale modeling of reaction rate by this new method because a previous plane two-dimensional simulation of this wake flame, using a wrinkling function and artificial flame thickening, had revealed discrepancies when compared with experiment. The present simulation is of the temporal evolution of a round wakelike flow at two Reynolds numbers, Re = 2000 and 10,000, based on wake defect velocity and wake diameter. A Fourier-spectral code has been used. The reaction is single-step and irreversible, and the rate follows an Arrhenius law. The reference simulation at the lower Reynolds number is fully resolved. At Re = 10,000, subgrid-scale contributions are significant. It was found that subgrid-scale modeling in the present simulation agrees more closely with unresolved subgrid-scale effects observed in experiment. Specifically, the highest contributions appeared in thin folded regions created by vortex convection. The wrinkling function approach had not selected subgrid-scale effects in these regions.

A new polymeric flame retardant additive for vinyl polymers

Diethyl allyl phosphate (DEAP) monomer has been synthesized, and characterized, using H-1 NMR and direct ionization mass spectrometric (DI-MS) techniques. It was free-radically polymerized to yield the poly(diethyl allyl phosphate) (PDEAP). The direct pyrolysis-mass spectrometric (DP-MS) analysis of the PDEAP revealed that it undergoes thermal degradation to yield mainly the monomer. Utility of PDEAP as a potent flame-retardant additive in polystyrene (PS) and poly(methyl methacrylate) (PMMA) has also been established.

Synthesis of nanocrystalline alpha-Al2O3 by ultrasonic flame pyrolysis

Nanocrystalline alpha-alumina was synthesized in an indigenously built ultrasonic flame pyrolysis (UFP) setup. This paper describes the technical aspects of the apparatus and particle formation in the flame. Ultrasonically atomized aluminium nitrate dissolved in methanol-water mixture was pyrolyzed in an oxy-propane flame for yielding nanocrystalline alpha-alumina. The formation of nanophase alumina was confirmed by powder XRD analysis. Scanning electron microscopy (SEM) analysis was carried out to study particulate morphology. (C) 2003 Elsevier Science Ltd. All rights reserved.

Tribology of ethylene-air diffusion flame soot under dry and lubricated contact conditions

Soot particles are generated in a flame caused by burning ethylene gas. The particles are collected thermophoretically at different locations of the flame. The particles are used to lubricate a steel/steel ball on flat reciprocating sliding contact, as a dry solid lubricant and also as suspended in hexadecane. Reciprocating contact is shown to establish a protective and low friction tribo-film. The friction correlates with the level of graphitic order of the soot, which is highest in the soot extracted from the mid-flame region and is low in the soot extracted from the flame root and flame tip regions. Micro-Raman spectroscopy of the tribo-film shows that the a priori graphitic order, the molecular carbon content of the soot and the graphitization of the film as brought about by tribology distinguish between the frictions of soot extracted from different regions of the flame, and differentiate the friction associated with dry tribology from that recorded under lubricated tribology.

Effect of Adiabatic Flame Temperature on Nano Alumina Powders during Solution Combustion Process

Nano ceramic alumina powders are synthesized by solution combustion synthesis using aluminium nitrate as oxidizer and urea as fuel with different fuel to oxidizer ratio. The variation of adiabatic flame temperatures are calculated theoretically for different fuel/oxidizer ratio according to thermodynamic concept and correlated with the observed flame (reaction) temperatures. A ``multi channel thermocouple setup connected to computer interfaced Keithley multi meter 2700'' is used to monitor the thermal events occurring during the process. The combustion products, characterized by XRD, show that the powders are composed of polycrystalline oxides with crystallite size of 32 to 52 nm. An interpretation based on maximum combustion temperature and the amount of gases produced during reaction for various fuel to oxide ratio has been proposed for the nature of combustion and its correlation with the characteristics of as-synthesized powder.

Dry tribology and nanomechanics of gaseous flame soot in comparison with carbon black and diesel soot

Ethylene gas is burnt and the carbon soot particles are thermophoretically collected using a home-built equipment where the fuel air injection and intervention into the 7.5-cm long flame are controlled using three small pneumatic cylinders and computer-driven controllers. The physical and mechanical properties and tribological performance of the collected soot are compared with those of carbon black and diesel soot. The crystalline structures of the nanometric particles generated in the flame, as revealed by high-resolution transmission electron studies, are shown to vary from the flame root to the exhaust. As the particle journeys upwards the flame, through a purely amorphous coagulated phase at the burner nozzle, it leads to a well-defined crystalline phase shell in the mid-flame zone and to a disordered phase consisting of randomly distributed short-range crystalline order at the exhaust. In the mid-flame region, a large shell of radial-columnar order surrounds a dense amorphous core. The hardness and wear resistance as well as friction coefficient of the soot extracted from this zone are low. The mechanical properties characteristics of this zone may be attributed to microcrystalline slip. Moving towards the exhaust, the slip is inhibited and there is an increase in hardness and friction compared to those in the mid-flame zone. This study of the comparison of flame soot to carbon black and diesel soot is further extended to suggest a rationale based on additional physico-chemical study using micro-Raman spectroscopy.

Time Series Classification for Online Tamil Handwritten Character Recognition – A Kernel Based Approach

In this paper, we consider the problem of time series classification. Using piecewise linear interpolation various novel kernels are obtained which can be used with Support vector machines for designing classifiers capable of deciding the class of a given time series. The approach is general and is applicable in many scenarios. We apply the method to the task of Online Tamil handwritten character recognition with promising results.

Premixed flame response to equivalence ratio fluctuations: Comparison between reduced order modeling and detailed computations

Combustion instability events in lean premixed combustion systems can cause spatio-temporal variations in unburnt mixture fuel/air ratio. This provides a driving mechanism for heat-release oscillations when they interact with the flame. Several Reduced Order Modelling (ROM) approaches to predict the characteristics of these oscillations have been developed in the past. The present paper compares results for flame describing function characteristics determined from a ROM approach based on the level-set method, with corresponding results from detailed, fully compressible reacting flow computations for the same two dimensional slot flame configuration. The comparison between these results is seen to be sensitive to small geometric differences in the shape of the nominally steady flame used in the two computations. When the results are corrected to account for these differences, describing function magnitudes are well predicted for frequencies lesser than and greater than a lower and upper cutoff respectively due to amplification of flame surface wrinkling by the convective Darrieus-Landau (DL) instability. However, good agreement in describing function phase predictions is seen as the ROM captures the transit time of wrinkles through the flame correctly. Also, good agreement is seen for both magnitude and phase of the flame response, for large forcing amplitudes, at frequencies where the DL instability has a minimal influence. Thus, the present ROM can predict flame response as long as the DL instability, caused by gas expansion at the flame front, does not significantly alter flame front perturbation amplitudes as they traverse the flame. (C) 2012 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

Efficient Methods for Robust Classification Under Uncertainty in Kernel Matrices

In this paper we study the problem of designing SVM classifiers when the kernel matrix, K, is affected by uncertainty. Specifically K is modeled as a positive affine combination of given positive semi definite kernels, with the coefficients ranging in a norm-bounded uncertainty set. We treat the problem using the Robust Optimization methodology. This reduces the uncertain SVM problem into a deterministic conic quadratic problem which can be solved in principle by a polynomial time Interior Point (IP) algorithm. However, for large-scale classification problems, IP methods become intractable and one has to resort to first-order gradient type methods. The strategy we use here is to reformulate the robust counterpart of the uncertain SVM problem as a saddle point problem and employ a special gradient scheme which works directly on the convex-concave saddle function. The algorithm is a simplified version of a general scheme due to Juditski and Nemirovski (2011). It achieves an O(1/T-2) reduction of the initial error after T iterations. A comprehensive empirical study on both synthetic data and real-world protein structure data sets show that the proposed formulations achieve the desired robustness, and the saddle point based algorithm outperforms the IP method significantly.