Block coordinate descent algorithms for large-scale sparse multiclass classification

Over the past decade, ? ₁ regularization has emerged as a powerful way to learn classifiers with implicit feature selection. More recently, mixed-norm (e.g., ? ₁/? ₂) regularization has been utilized as a way to select entire groups of features. In this paper, we propose a novel direct multiclass formulation specifically designed for large-scale and high-dimensional problems such as document classification. Based on a multiclass extension of the squared hinge loss, our formulation employs ? ₁/? ₂ regularization so as to force weights corresponding to the same features to be zero across all classes, resulting in compact and fast-to-evaluate multiclass models. For optimization, we employ two globally-convergent variants of block coordinate descent, one with line search (Tseng and Yun in Math. Program. 117:387–423, 2009) and the other without (Richtárik and Taká? in Math. Program. 1–38, 2012a; Tech. Rep. arXiv:1212.0873, 2012b). We present the two variants in a unified manner and develop the core components needed to efficiently solve our formulation. The end result is a couple of block coordinate descent algorithms specifically tailored to our multiclass formulation. Experimentally, we show that block coordinate descent performs favorably compared to other solvers such as FOBOS, FISTA and SpaRSA. Furthermore, we show that our formulation obtains very compact multiclass models and outperforms ? ₁/? ₂-regularized multiclass logistic regression in terms of training speed, while achieving comparable test accuracy.     

Construction and development of a time-resolved x-ray magnetic circular dichroism-photoelectron emission microscopy system using femtosecond laser pulses at BL25SU SPring-8

A femtosecond pulsed laser system has been installed at the BL25SU soft x-ray beamline at SPring-8 for time-resolved pump-probe experiments with synchronization of the laser pulses to the circularly polarized x-ray pulses. There are four different apparatuses situated at the beamline; for photoemission spectroscopy, two-dimensional display photoelectron diffraction, x-ray magnetic circular dichroism (XMCD) with electromagnetic coils, and photoelectron emission microscopy (PEEM). All four can be used for time-resolved experiments, and preliminary investigations have been carried out using the PEEM apparatus to observe magnetization dynamics in combination with XMCD. In this article, we describe the details of the stroboscopic pump-probe XMCD-PEEM experiment, and present preliminary data. The repetition rate of the laser pulses is set using a pulse selector to match the single bunches of SPring-8's hybrid filling pattern, which consists of several single bunches and a continuous bunch train. Electrons ejected during the bunch train, which do not provide time-resolved signal, are eliminated by periodically reducing the channel plate voltage using a custom-built power supply. The pulsed laser is used to create 300 ps long magnetic field pulses, which cause magnetic excitations in micron-sized magnetic elements which contain magnetic vortex structures. The observed frequency of the motion is consistent with previously reported observations and simulations.     

Effects of entraining velocity of lubricant and sliding velocity on friction behavior in stainless steel sheet rolling

A series of experiments was carried out using a rolling-type tribometer to investigate the effects on friction behavior of the entraining velocity of the lubricant at the inlet to the contact zone (V) and sliding velocity during deformation (ΔV). Experiments with stainless steel sheets of two different surface roughnesses showed that the variations in the friction coefficient with entraining velocity V and sliding velocity ΔV are largely dependent on the initial surface texture of the workpiece. For a smooth workpiece, the friction coefficient decreases with increasing sliding velocity ΔV but keeps almost constant with increasing entraining velocity V. However, for a rough workpiece, the friction coefficient initially decreases slowly and increases largely with increasing sliding velocity ΔV or decreasing entraining velocity V. Observation of the rolled surface for a smooth workpiece shows that, with increasing entraining velocity V, the slip band becomes more marked, and with increasing sliding velocity ΔV, the rubbed portions become more conspicuous. For a rough workpiece, galling occurs at high sliding velocity ΔV. The critical condition for galling outbreak is shown on the V-ΔV graph. The galling outbreak process is observed by interrupting the rolling process.     

Prediction of Bending Fatigue Life of Cracked Out-of-Plane Gusset Joint Repaired by CFRP Plates

Carbon fiber reinforced polymer (CFRP), plates bonding repair method is one of the simple repair methods for cracked steel structures. In this study, the influence of width of CFRP plates on bending fatigue life of out-of-plane gusset joint strengthened with CFRP plates was investigated from the experimental and numerical point of view. In the bending fatigue test of cracked out-of-plane gusset joint strengthened with CFRP plates, the effect of width of CFRP plates on crack growth life was clarified experimentally. Namely, it was revealed that the crack growth life becomes larger with increasing the width of CFRP plates. In the numerical approach, the stress intensity factor (SIF) at the surface point of a semi-elliptical surface crack was estimated based on the linear fracture mechanics. Furthermore, the extended fatigue life of cracked out-of-plane gusset joint strengthened with CFRP plates was evaluated by using the estimated SIF at the surface point and the empirical formula of the aspect ratio of semi-elliptical crack. As the results of numerical analysis, the estimated fatigue life of the specimen strengthened with CFRP plates showed the good agreement with the test results.     

Morphological evolution of NiO-ScSZ composite in a high-temperature reducing atmosphere

The microstructural change in composite of nickel oxide and scandia-stabilized zirconia (NiO-ScSZ) under a reducing atmosphere was observed by scanning electron microscopy (SEM). The morphological transformation was noticeable after high-temperature treatment with the formation of two peculiar microstructures; i.e., fibrous zirconia and metallic nickel with wrinkled surface. It was suggested that partial reduction of the nickel species dissolved in ScSZ lattice triggered the formation of these characteristic morphologies. The growth of fibrous zirconia appeared to be promoted via interfacial reaction between the metallic Ni particles and the zirconia phase. The agglomeration of metallic nickel proceeded by the reduction at high temperatures and then the surface was transformed to the wrinkled morphology. The unique wrinkled pattern was often observed for the large agglomerated nickel particles. This drastic change in Ni-ScSZ microstructure upon the high-temperature reduction degraded the electrochemical performance of solid oxide fuel cells.     

Synthesis and Properties of Functionalized [6] Helicenes by the Thermal Domino Radical Cycloaromatization of Acyclic Polyynes

Non-conjugated aromatic hexaynes and heptaynes undergo thermal domino radical cycloaromatization under mild conditions to yield indenol ring-fused fluoranthene and indenol ring-fused benzo[a]rubicene skeletons with helicity. The structure and thermal racemization of [6]helicene derivatives (12 and 13) are also described.     

Preparation of SiO2-B2O3 glass forPANDA optical fibre from colloidal sol-gel process

SiO₂-B₂O₃ glass for the PANDA optical fibre is prepared from the colloidal sol-gel process. The preparation of SiO₂-B₂O₃ glass is through the gellation of fumed silica dispersed in water containing boric acid. Two processes to dry the gel are examined, and freeze-drying is selected as the better drying method. The process produces a dried gel without cracks and with a uniform distribution of boron oxide concentration. The dried gel is sintered into glass     

Calorimetric study of mutant human lysozymes with partially introduced Ca2+ binding sites and its efficient refolding system from inclusion bodies

During the process of evolution, ancestral lysozymes evolved into calcium-binding lysozymes by acquiring three critical aspartate residues at positions 86, 91 and 92. To investigate the process of the acquisition of calcium-binding ability, two of the aspartates were partially introduced into human lysozyme at positions 86, 91 and 92. These mutants (HLQ86D, HLA92D and HLQ86D/D91Q/A92D), having two critical aspartates in calcium-binding sites, were expressed in Escherichia coli as non-active inclusion bodies. For the preparation of lysozyme samples, a refolding system using thioredoxin was established. This system allowed for effective refolding of wild-type and mutant lysozymes, and 100% of activity was recovered within 4 days. The calcium ion dependence of the melting temperature (Tm) of wild-type and mutant lysozymes was investigated by differential scanning calorimetry at pH 4.5. The Tm values of wild-type, HLQ86D and HLA92D mutants were not dependent on calcium ion concentration. However, the Tm of HLQ86D/D91Q/A92D was 4 degrees higher in the presence of 50 mM CaCl2 than in its absence, and the calcium-binding constant of this mutant was estimated to be 2.25(+/-0.25)x10(2) M(-1) at pH 4.5. Moreover, the calcium-binding ability of this mutant was confirmed by the result using Sephadex G-25 gel chromatography. These results indicate that it is indispensable to have at least two aspartates at positions 86 and 92 for acquisition of calcium-binding ability. The process of the acquisition of calcium-binding site during evolution of calcium-binding lysozyme is discussed.