Korea University

About: Korea University is a education organization based out in Seoul, South Korea. It is known for research contribution in the topics: Population & Catalysis. The organization has 39756 authors who have published 82424 publications receiving 1860927 citations. The organization is also known as: Bosung College & Bosung Professional College.

"What is relevant in a text document?": An interpretable machine learning approach

Abstract: Text documents can be described by a number of abstract concepts such as semantic category, writing style, or sentiment. Machine learning (ML) models have been trained to automatically map documents to these abstract concepts, allowing to annotate very large text collections, more than could be processed by a human in a lifetime. Besides predicting the text’s category very accurately, it is also highly desirable to understand how and why the categorization process takes place. In this paper, we demonstrate that such understanding can be achieved by tracing the classification decision back to individual words using layer-wise relevance propagation (LRP), a recently developed technique for explaining predictions of complex non-linear classifiers. We train two word-based ML models, a convolutional neural network (CNN) and a bag-of-words SVM classifier, on a topic categorization task and adapt the LRP method to decompose the predictions of these models onto words. Resulting scores indicate how much individual words contribute to the overall classification decision. This enables one to distill relevant information from text documents without an explicit semantic information extraction step. We further use the word-wise relevance scores for generating novel vector-based document representations which capture semantic information. Based on these document vectors, we introduce a measure of model explanatory power and show that, although the SVM and CNN models perform similarly in terms of classification accuracy, the latter exhibits a higher level of explainability which makes it more comprehensible for humans and potentially more useful for other applications.

Composition and major sources of organic compounds of aerosol particulate matter sampled during the ACE-Asia campaign

Abstract: [1] The organic compound tracers of atmospheric particulate matter, as well as organic carbon (OC) and elemental carbon (EC), have been characterized for samples acquired during the Asian Pacific Regional Aerosol Characterization Experiment (ACE-Asia) from Gosan, Jeju Island, Korea, from Sapporo, Japan, and from Chichi-jima Island in the western North Pacific, as well as on the National Oceanic and Atmospheric Administration R/V Ronald H. Brown. Total extracts were analyzed by gas chromatography–mass spectrometry to determine both polar and aliphatic compounds. Total particles, organic matter, and lipid and saccharide compounds were high during the Asian dust episode (early April 2001) compared to levels at other times. The organic matter can be apportioned to seven emission sources and to significant oxidation-producing secondary products during long-range transport. Terrestrial natural background compounds are vascular plant wax lipids derived from direct emission and as part of desert sand dust. Fossil fuel utilization is obvious and derives from petroleum product and coal combustion emissions. Saccharides are a major polar (water-soluble) carbonaceous fraction derived from soil resuspension (agricultural activities). Biomass-burning smoke is evident in all samples and seasons. It contributes up to 13% of the total compound mass as water-soluble constituents. Burning of refuse is another source of organic particles. Varying levels of marine-derived lipids are superimposed during aerosol transport over the ocean. Secondary oxidation products increase with increasing transport distance and time. The ACE-Asia aerosols are composed not only of desert dust but also of soil dust, smoke from biomass and refuse burning, and emissions from fossil fuel use in urban areas.

Adaptive Fuzzy Control of Nonlinear Systems With Unmodeled Dynamics and Input Saturation Using Small-Gain Approach

Abstract: This paper investigates the problem of adaptive fuzzy state-feedback control for a category of single-input and single-output nonlinear systems in nonstrict-feedback form. Unmodeled dynamics and input constraint are considered in the system. Fuzzy logic systems are employed to identify unknown nonlinear characteristics existing in systems. An appropriate Lyapunov function is chosen to ensure unmodeled dynamics to be input-to-state practically stable. A smooth function is introduced to tackle input saturation. In order to overcome the difficulty of controller design for nonstrict-feedback system in backstepping design process, a variables separation method is introduced. Moreover, based on small-gain technique, an adaptive fuzzy controller is designed to guarantee all the signals of the resulting closed-loop system to be bounded. Finally, two illustrative examples are given to validate the effectiveness of the new design techniques.

Observation of sequential Υ suppression in PbPb collisions

Abstract: The suppression of the individual Υ(nS) states in PbPb collisions with respect to their yields in pp data has been measured. The PbPb and pp data sets used in the analysis correspond to integrated luminosities of 150 μb^(-1) and 230 nb^(-1), respectively, collected in 2011 by the CMS experiment at the LHC, at a center-of-mass energy per nucleon pair of 2.76 TeV. The Υ(nS) yields are measured from the dimuon invariant mass spectra. The suppression of the Υ(nS) yields in PbPb relative to the yields in pp, scaled by the number of nucleon-nucleon collisions, R_(AA), is measured as a function of the collision centrality. Integrated over centrality, the R_(AA) values are 0.56±0.08(stat)±0.07(syst), 0.12±0.04(stat)±0.02(syst), and lower than 0.10 (at 95% confidence level), for the Υ(1S), Υ(2S), and Υ(3S) states, respectively. The results demonstrate the sequential suppression of the Υ(nS) states in PbPb collisions at LHC energies.

Ultrafine SnO2 nanoparticle loading onto reduced graphene oxide as anodes for sodium-ion batteries with superior rate and cycling performances

Abstract: A structured SnO2–reduced graphene oxide (RGO) nanocomposite has been synthesized with SnO2 nanoparticles (∼5 nm) anchored on a RGO framework. It has been successfully applied as an anode material in sodium-ion batteries. The electrode delivers a reversible Na-storage capacity of 330 mA h g−1 with an outstanding capacity retention of 81.3% over 150 cycles. Moreover, it possesses a relatively good rate capability, exhibiting a capacity retention of 25.8% at high rate (1000 mA h g−1). With its combined advantages of low cost and environmental benignity, the SnO2–RGO nanocomposite would be a promising anode for Na-ion batteries.