AUBER: Automated BERT regularization.

How can we effectively regularize BERT? Although BERT proves its effectiveness in various NLP tasks, it often overfits when there are only a small number of training instances. A promising direction to regularize BERT is based on pruning its attention heads with a proxy score for head importance. However, these methods are usually suboptimal since they resort to arbitrarily determined numbers of attention heads to be pruned and do not directly aim for the performance enhancement.

Physicochemical properties of chars at different treatment temperatures.

In this study, the physicochemical properties of the char of Indonesian SM coal following heat treatment at various temperatures were evaluated using X-ray photoelectron spectroscopy (XPS), Fourier-transform infrared spectroscopy (FT-IR), scanning electron microscopy (SEM), and morphological and specific surface area analysis. Based on these analyses, heat treatment of coal was determined to be the most effective in increasing the coal rank. In the XPS analysis, the C-O and C-O-C groups and quaternary-N species were found to be of a lower grade coal when the pretreatment temperature decreased, meanwhile the C-C group and pyridinic species increased.

Characteristics of fundamental combustion and NOx emission using various rank coals.

Eight types of coals of different rank were selected and their fundamental combustion characteristics were examined along with the conversion of volatile nitrogen (N) to nitrogen oxides (NOx)/fuel N to NOx. The activation energy, onset temperature, and burnout temperature were obtained from the differential thermogravimetry curve and Arrhenius plot, which were derived through thermo-gravimetric analysis. In addition, to derive the combustion of volatile N to NOx/fuel N to NOx, the coal sample, which was pretreated at various temperatures, was burned, and the results were compared with previously derived fundamental combustion characteristics.