Learning with symmetric label noise: The importance of being unhinged

Convex potential minimisation is the de facto approach to binary classification. However, Long and Servedio [2008] proved that under symmetric label noise (SLN), minimisation of any convex potential over a linear function class can result in classification performance equivalent to random guessing. This ostensibly shows that convex losses are not SLN-robust. In this paper, we propose a convex, classification-calibrated loss and prove that it is SLN-robust. The loss avoids the Long and Servedio [2008] result by virtue of being negatively unbounded. The loss is a modification of the hinge loss, where one does not clamp at zero; hence, we call it the unhinged loss. We show that the optimal unhinged solution is equivalent to that of a strongly regularised SVM, and is the limiting solution for any convex potential; this implies that strong l2 regularisation makes most standard learners SLN-robust. Experiments confirm the unhinged loss’ SLN-robustness.

General properties of electrochemical capacitors

The use of capacitors for electrical energy storage actually predates the invention of the battery. Alessandro Volta is attributed with the invention of the battery in 1800, where he first describes a battery as an assembly of plates of two different materials (such as copper and zinc) placed in an alternating stack and separated by paper soaked in brine or vinegar [1]. Accordingly, this device was referred to as Volta’s pile and formed the basis of subsequent revolutionary research and discoveries on the chemical origin of electricity. Before the advent of Volta’s pile, however, eighteenth century researchers relied on the use of Leyden jars as a source of electrical energy. Built in the mid-1700s at the University of Leyden in Holland, a Leyden jar is an early capacitor consisting of a glass jar coated inside and outside with a thin layer of silver foil [2, 3]. With the outer foil being grounded, the inner foil could be charged with an electrostatic generator, or a source of static electricity, and could produce a strong electrical discharge from a small and comparatively simple device.

Evaluate traffic noise level based on traffic microsimulation combined with a refined classic noise prediction method

In this paper, a refined classic noise prediction method based on the VISSIM and FHWA noise prediction model is formulated to analyze the sound level contributed by traffic on the Nanjing Lukou airport connecting freeway before and after widening. The aim of this research is to (i) assess the traffic noise impact on the Nanjing University of Aeronautics and Astronautics (NUAA) campus before and after freeway widening, (ii) compare the prediction results with field data to test the accuracy of this method, (iii) analyze the relationship between traffic characteristics and sound level. The results indicate that the mean difference between model predictions and field measurements is acceptable. The traffic composition impact study indicates that buses (including mid-sized trucks) and heavy goods vehicles contribute a significant proportion of total noise power despite their low traffic volume. In addition, speed analysis offers an explanation for the minor differences in noise level across time periods. Future work will aim at reducing model error, by focusing on noise barrier analysis using the FEM/BEM method and modifying the vehicle noise emission equation by conducting field experimentation.

Electrochemical properties and intermediate-temperature fuel cell performance of dense yttrium-doped barium zirconate with calcium addition

Although BaZr 0.8Y 0.2O 3-δ(BZY) possesses large bulk proton conductivity and excellent chemical stability, its poor sinterability and grain boundaries block proton conduction. In this work, the effect of Ca as a co-dopant and as a sintering aid (as CaO), on the sinterability, proton conductivity, and fuel cell performance of BZY was investigated. The addition of 4 mol% CaO significantly improved the BZY sinterability: BZY pellets with densities of 92.7% and 97.5% with respect to the theoretical density were obtained after sintering at 1500°C and 1600°C, respectively. The improved BZY sinterability by CaO addition resulted also in a large proton conductivity; at 600°C, the total conductivity of BZY-CaO was 2.14 × 10 -3 S/cm, in wet Ar. Anode-supported fuel cells with 25 μm-thick BZY-CaO electrolyte membranes were fabricated by a dual-layer co-firing technique. The peak power density of the fuel cell with a BZY-Ni/BZY-4CaO/BZY-LSCF (La 0.6Sr 0.4Fe 0.8Co 0.2O 3-δ) configuration was 141 mW/cm 2 at 700°C, several times larger than the reported values of BZY electrolyte membrane fuel cells sintered with the addition of CuO or ZnO, demonstrating promising features for practical fuel cell applications.

Sinteractive anodic powders improve densification and electrochemical properties of BaZr0.8Y0.2O3-δ electrolyte films for anode-supported solid oxide fuel cells

The difficult sintering of BaZr0.8Y0.2O 3-δ (BZY20) powders makes the fabrication of anode-supported BZY20 electrolyte films complex. Dense BZY20 membranes were successfully fabricated on anode substrates made of sinteractive NiO-BZY20 powders, prepared by a combustion method. With respect to traditional anode substrates made of powders prepared by mechanical mixing, the anode substrates made of the wet-chemically synthesized composite NiO-BZY20 powders significantly promoted the densification of BZY20 membranes: dense BZY20 films were obtained after co-pressing and co-firing at 1300 °C, a much lower temperature than those usually needed for densifying BZY20 membranes. Improved electrochemical performance was also observed: the supported BZY20 films maintained a high proton conductivity, up to 5.4 × 10-3 S cm-1 at 700 °C. Moreover, an anode-supported fuel cell with a 30 m thick BZY20 electrolyte film fabricated at 1400 °C on the anode made of the wet-chemically synthesized NiO-BZY20 powder showed a peak power density of 172 mW cm-2 at 700 °C, using La0.6Sr0.4Co 0.2Fe0.8O3-δ-BaZr0.7Y 0.2Pr0.1O3-δ as the cathode material, with a remarkable performance for proton-conducting solid oxide fuel cell (SOFC) applications.

Real-time kinematic positioning over long baselines using triple-frequency BeiDou signals

The BeiDou system is the first global navigation satellite system in which all satellites transmit triple-frequency signals that can provide the positioning, navigation, and timing independently. A benefit of triple-frequency signals is that more useful combinations can be formed, including some extrawide-lane combinations whose ambiguities can generally be instantaneously fixed without distance restriction, although the narrow-lane ambiguity resolution (NL AR) still depends on the interreceiver distance or requires a long time to achieve. In this paper, we synthetically study decimeter and centimeter kinematic positioning using BeiDou triple-frequency signals. It starts with AR of two extrawide-lane signals based on the ionosphere-free or ionosphere-reduced geometry-free model. For decimeter positioning, one can immediately use two ambiguity-fixed extrawide-lane observations without pursuing NL AR. To achieve higher accuracy, NL AR is the necessary next step. Despite the fact that long-baseline NL AR is still challenging, some NL ambiguities can indeed be fixed with high reliability. Partial AR for NL signals is acceptable, because as long as some ambiguities for NL signals are fixed, positioning accuracy will be certainly improved.With accumulation of observations, more and more NL ambiguities are fixed and the positioning accuracy continues to improve. An efficient Kalman-filtering system is established to implement the whole process. The formulated system is flexible, since the additional constraints can be easily applied to enhance the model's strength. Numerical results from a set of real triple-frequency BeiDou data on a 50 km baseline show that decimeter positioning is achievable instantaneously.With only five data epochs, 84% of NL ambiguities can be fixed so that the real-time kinematic accuracies are 4.5, 2.5, and 16 cm for north, east, and height components (respectively), while with 10 data epochs more than 90% of NL ambiguities are fixed, and the rea- -time kinematic solutions are improved to centimeter level for all three coordinate components.

Assessment of code bias variations of BDS triple-frequency signals and their impacts on ambiguity resolution for long baselines

Carrier phase ambiguity resolution over long baselines is challenging in BDS data processing. This is partially due to the variations of the hardware biases in BDS code signals and its dependence on elevation angles. We present an assessment of satellite-induced code bias variations in BDS triple-frequency signals and the ambiguity resolutions procedures involving both geometry-free and geometry-based models. First, since the elevation of a GEO satellite remains unchanged, we propose to model the single-differenced fractional cycle bias with widespread ground stations. Second, the effects of code bias variations induced by GEO, IGSO and MEO satellites on ambiguity resolution of extra-wide-lane, wide-lane and narrow-lane combinations are analyzed. Third, together with the IGSO and MEO code bias variations models, the effects of code bias variations on ambiguity resolution are examined using 30-day data collected over the baselines ranging from 500 to 2600 km in 2014. The results suggest that although the effect of code bias variations on the extra-wide-lane integer solution is almost ignorable due to its long wavelength, the wide-lane integer solutions are rather sensitive to the code bias variations. Wide-lane ambiguity resolution success rates are evidently improved when code bias variations are corrected. However, the improvement of narrow-lane ambiguity resolution is not obvious since it is based on geometry-based model and there is only an indirect impact on the narrow-lane ambiguity solutions.

Geometry-free stochastic analysis of BDS triple frequency signals

The paper presents a geometry-free approach to assess the variation of covariance matrices of undifferenced triple frequency GNSS measurements and its impact on positioning solutions. Four independent geometryfree/ ionosphere-free (GFIF) models formed from original triple-frequency code and phase signals allow for effective computation of variance-covariance matrices using real data. Variance Component Estimation (VCE) algorithms are implemented to obtain the covariance matrices for three pseudorange and three carrier-phase signals epoch-by-epoch. Covariance results from the triple frequency Beidou System (BDS) and GPS data sets demonstrate that the estimated standard deviation varies in consistence with the amplitude of actual GFIF error time series. The single point positioning (SPP) results from BDS ionosphere-free measurements at four MGEX stations demonstrate an improvement of up to about 50% in Up direction relative to the results based on a mean square statistics. Additionally, a more extensive SPP analysis at 95 global MGEX stations based on GPS ionosphere-free measurements shows an average improvement of about 10% relative to the traditional results. This finding provides a preliminary confirmation that adequate consideration of the variation of covariance leads to the improvement of GNSS state solutions.

Wide area kinematic positioning with BeiDou triple frequency signals

A key challenge of wide area kinematic positioning is to overcome the effects of the varying hardware biases in code signals of the BeiDou system. Based on three geometryfree/ionosphere-free combinations, the elevation-dependent code biases are modelled for all BeiDou satellites. Results from the data sets of 30-day for 5 baselines of 533 to 2545 km demonstrate that the wide-lane (WL) integer-fixing success rates of 98% to 100% can be achieved within 25 min. Under the condition of HDOP of less than 2, the overall RMS statistics show that ionospheric-free WL single-epoch solutions achieve 24 to 50 cm in the horizontal direction. Smoothing processing over the moving window of 20 min reduces the RMS values by a factor of about 2. Considering distance-independent nature, the above results show the potential that reliable and high precision positioning services could be provided in a wide area based on a sparsely distributed ground network.

Nano-confined synthesis of highly ordered mesoporous carbon and its performance as electrode material for electrochemical behavior of riboflavin (vitamin B2) and dopamine

Highly ordered mesoporous carbon (MC) has been synthesized from sucrose, a non-toxic and costeffective source of carbon. X-ray diffraction, N2 adsorption–desorption isotherm and transmission electron micrograph (TEM) were used to characterize the MC. The XRD patterns show the formation of highly ordered mesoporous structures of SBA15 and mesoporous carbon. The N2 adsorptiondesorption isotherms suggest that the MC exhibits a narrow pore-size distribution with high surface area of 1559 m2/g. The potential application of MC as a novel electrode material was investigated using cyclic voltammetry for riboflavin (vitamin B2) and dopamine. MC-modified glassy carbon electrode (MC/GC) shows increase in peak current compared to GC electrode in potassium ferricyanide which clearly suggest that MC/GC possesses larger electrode area (1.8 fold) compared with bare GC electrode. The electrocatalytic behavior of MC/GC was investigated towards the oxidation of riboflavin (vitamin B2) and dopamine using cyclic voltammetry which show larger oxidation current compared to unmodified electrode and thus MC/GC may have the potential to be used as a chemically modified electrode.