UN and many countries are calling for urgent measures because air pollution caused by air pollutants seriously threatens human health and causes premature death. Many investigations and studies have revealed that PM2.5 is a main pollutant that destroys the environment of ecosystems and harms human health, and that air pollutants containing PM 2.5 cause various diseases such as respiratory, lung and heart diseases.

In the air pollution concentration prediction task, it was reported that the artificial neural network (ANN) model improves prediction performance while avoiding the complexity and annoyance of modeling, compared to the deterministic method which has lack in representing the heterogeneity and nonlinearity of many factors related to pollutant formation.

Therefore, many researchers have begun to apply artificial neural networks extensively to air pollution prediction, and since then they have ensured the accurate prediction performance by combining deep neural networks with several other optimization techniques that have shown dramatic effectiveness in the prediction of large time series data. In a word, their research results show that deep learning models significantly improve their performance when combined with several other effective analytical methods on time series data rather than being used as a single model.

Based on this analysis, Pak Un Jin, a researcher at the Faculty of Automatics, proposed a hybrid model consisting of convolutional neural network (CNN)-long short-term (LSTM) with multifractal detrended fluctuation analysis (MF-DFA) for air pollution and meteorological time series data, and used it to predict the next day’s 24-h average PM2.5 concentration in Beijing City.

The comparison of the performance indexes of the proposed model with MLP and LSTM models showed that the proposed model provides higher prediction accuracy.

For further details, please refer to his paper “A deep learning approach via multifractal detrended fluctuation analysis for PM2.5 prediction” in “Journal of Atmospheric and Solar-Terrestrial Physics” (SCI).

The evacuated tube solar collector (ETSC) shows superior performance over the flat-plate collector (FPC) as it can reduce convective and radiation heat loss, and the thermal efficiency of ETSC has a linear relationship with the difference between the fluid inlet temperature and the ambient temperature divided by the solar radiation incident on the collector.

Under steady-state operating conditions, the useful output power of FPC is a function of many parameters, and one of them is the overall heat loss coefficient of FPC.

WGETSC (water-in-glass evacuated tube solar collector), unlike FPC, typically consists of 16-28 flooded solar tubes in direct connection to a horizontal water tank, so it is difficult to use the single overall heat loss coefficient as in FPC.

Having considered the solar tube and the water tank respectively in WGETSC, Jong Hyon Il, a section head at the Faculty of Thermal Engineering, derived a formula to determine the thermal performance of WGETSC and compared the calculated value with measured one.

The mean relative error between the measured and calculated values was 9.8%.

For more information, you can refer to his paper “On evaluating thermal performance of water-in-glass evacuated tube solar collector” in “International Journal of Research in Engineering”.

As face is rather invariable among human biometric features, face recognition is considered as the most important biometric identification task. Face recognition has been widely used for user authentication in security systems such as electronic payment systems as a technology to search faces from facial images from photographs or videos and therefore, it has been studied for years.

Over the past decade, the mainstream of face recognition has been based on deep learning and it has been developed to a much higher level than average human ability. Nevertheless, face recognition is still far from “perfect”.

Recently, face recognition systems using convolutional neural networks are considered as the best methods among existing face recognition systems. Face recognition network models that have been developed and proved to be superior in performance cannot perform real-time face recognition in devices with constrained computational resources such as low base computers or mobile phones because their structure is very complex and they need a large amount of computation. What is more, reducing the number of layers continuously to reduce computational burden affects recognition performance.

In GhostFaceNets, they improved the trade-off between speed and accuracy by performing the attention operation using a DFC (decoupled fully-connected) attention. However, the DFC attention has limitations in capturing wide spatial information, which may lead to the degradation of recognition performance.

Jo Kwang Chol, a researcher at the Institute of Information Technology, has designed a network structure with low computational cost and improved performance by combining the self-attention module with the extended Ghost module based on the backbone of GhostFaceNets, and verified its accuracy using international standard databases.

The results showed that the proposed network model brings significant improvement in face recognition performance with 99.74% in LFW and 97.7% in AgeDB-30 and that with 42 MFLOP, it can support stable real-time face recognition in embedded devices.

For more details, you can refer to his paper “GhostFormerNet: A Lightweight Face Recognition Method based on Extended Ghost Module and Self-Attention” in “2025 International Conference on Graphics and Signal Processing”.