Search

In Korean text these days, the use of English words with or without phonetic translations are growing at a high speed. To make matters worse, the Korean transliteration of an English word may vary greatly. The mixed use of English words and their various transliterations in the same document or document collection may cause severe word mismatch problems in Korean information retrieval. There are two possible approaches to tackle this problem: transliteration and backtransliteration method. We argue that our newly proposed transliteration approach is more advantageous for the resolution of the word mismatch problem than the previously proposed back-transliteration approach. Our information retrieval experiment results support this argument. [ABSTRACT FROM AUTHOR]

• Form a large facial expression dataset by combining public datasets. • Propose a combined center dispersion loss function to achieve high accuracy. • Propose an incremental cosine annealing to obtain a diversely trained ensemble. We propose a combined center dispersion loss function to reduce the intra-class variations and inter-class similarities of facial expression datasets and achieve high accuracy in facial expression recognition. Because of the lack of data, we strategically combine four publicly available facial expression datasets for training. Moreover, we propose an incremental cosine annealing method for deploying multiple models trained with incremental learning rates and ensemble predictions for achieving better accuracy. This method also reduces the computational cost and yields ensemble predictions of varied models, instead of similar models, that are trained with the same learning rates. We train our methods using the VGGFace network and achieve an accuracy of 74.71% on the FER2013 test set. Image, graphical abstract [ABSTRACT FROM AUTHOR]

Parallel structure is a way to factor out common constituents in the expressions, which makes an effect of simplification of expressions. The complexity can be greatly reduced at the phase of sentence parsing by identifying such boundaries of parallel structure. In this paper, we propose a probabilistic model to identify parallel cores (corresponding constituents) as well as boundaries of parallel noun phrases conjoined by "wa/gwa" (conjunctive particle in Korean). It is based on the idea of swapping constituents, utilizing symmetry (two or more identical constituents are repeated) and reversibility (the order of constituents is changeable) in parallel structure. The probabilities are calculated from (unlabelled) corpus with parallel structures, which is an advantage over the approaches trained with labeled corpus. Our model, moreover, is not dependent on languages. It is also shown that the semantic features of the modifiers around parallel noun phrase and the patterns among words can be utilized further to correct the boundaries identified by the swapping model. Experiment shows that our probabilistic swapping model performs much better than symmetry-based model and machine learning based approaches. [ABSTRACT FROM AUTHOR]