We present EDA: easy data augmentation techniques for boosting performance on text classification tasks. EDA consists of four simple but powerful operations: synonym replacement, random insertion, random swap, and random deletion. On five text classification tasks, we show that EDA improves performance for both convolutional and recurrent neural networks. EDA demonstrates particularly strong results for smaller datasets; on average, across five datasets, training with EDA while using only 50% of the available training set achieved the same accuracy as normal training with all available data. We also performed extensive ablation studies and suggest parameters for practical use.

No abstract

BackgroundEffective treatment of HIV since 1996 has reduced morbidity and mortality through virologic suppression. Combination antiretroviral therapy (cART) has been recognized as key to the prevention of drug resistance and the transmission of infection. We used eighteen years of virologic outcomes in a long-standing cohort of women to describe longitudinal viral load trajectories; and examine factors associated with sustained viremia and mortality.MethodsWe analyzed data from DC WIHS women with > four semiannual visits using a group-based logistic trajectory analysis approach to identify patterns of HIV RNA detection (>80 copies/mL or lower assay limit, and >1000 copies/mL). We verified findings using cumulative viral load suppression-years, explored group characteristics using generalized linear modeling with generalized estimating equations for repeated measures, and examined survival using the Kaplan-Meier and Cox proportional hazard analyses.Results329 women contributed 6633 visits between 1994 and 2012 and demonstrated high, moderate, and low probability patterns of HIV RNA detection (>80 copies/mL) in 40.7, 35.6, and 23.7 % of participant visits, respectively. Analysis of cumulative years of viral load suppression supported these observations. Kaplan-Meier survival analysis demonstrated high mortality of 31.1 % with sustained viremia, but no significant difference in mortality between intermittent viremia and non-viremia patterns, 6.9 and 4.9 % respectively. Mortality was associated with higher age, lower CD4+ T lymphocyte count, and sustained viremia by Cox multivariate analysis.ConclusionsThis ecologic study demonstrates the effectiveness of viral suppression, and conversely the association between viremia and mortality. In community delivery of cART for HIV care, distinct patterns of sustained viremia, intermittent viremia, and non-viremia were identified over nearly 18 years in the DC WIHS, capturing the dynamics and complexity of sustaining long-term HIV care. Persistent viremia was associated with lower CD4s and mortality, but surprisingly mortality was not different between continuous suppression and intermittent viremia. Classification of long-term virologic patterns such as these observed HIV treatment “careers” may provide a suitable framework to identify modifiable factors associated with treatment resilience and failure. Both individual and population interventions are needed to reduce transmission, prevent the emergence of drug resistance, and improve outcomes of community ART programs.

In this study, segregation roasting and magnetic separation are used to extract nickel from a garnierite laterite ore. The garnierite laterite ore containing 0.72% Ni, 0.029% Co, 8.65% Fe, 29.66% MgO, and 37.86% SiO 2 was collected in the Mojiang area of China. Garnierite was the Ni-bearing mineral; the other main minerals were potash feldspar, forsterite, tremolite, halloysite, quartz, and kaolinite in the garnierite laterite ore. The iron phase transformations show that nickel is transformed from (Ni,Mg)O·SiO 2 ·nH 2 O to a new nickel mineral phase dominated by [Ni]Fe solid solution; and iron changed from Fe 2 O 3 and FeOOH to a new iron mineral phase dominated by metal Fe and Fe 3 O 4 after segregation roasting. Ferronickel concentrate with Ni of 16.16%, Fe of 73.67%, and nickel recovery of 90.33% was obtained under the comprehensive conditions used: A roasting temperature of 1100 • C, a roasting time of 90 min, a calcium chloride dosage of 15%, an iron concentrate dosage of 30%, a coke dosage of 15%, a coke size of −1 + 0.5 mm, a magnetic separation grinding fineness of <45 µm occupying 90%, and a magnetic separation magnetic field intensity of H = 0.10 T. The main minerals in ferronickel concentrate are Fe, [Ni]Fe, Fe 3 O 4 , and a small amount of gangue minerals, such as CaO·SiO 2 and CaO·Al 2 O 3 ·SiO 2 .

No abstract