Cocrystals are of much interest in industrial application as well as academic research, and screening of suitable coformers for active pharmaceutical ingredients is the most crucial and challenging step in cocrystal development. Recently, machine learning techniques are attracting researchers in many fields including pharmaceutical research such as quantitative structure-activity/property relationship. In this paper, we develop machine learning models to predict cocrystal formation. We extract descriptor values from simplified molecular-input line-entry system (SMILES) of compounds and compare the machine learning models by experiments with our collected data of 1476 instances. As a result, we found that artificial neural network shows great potential as it has the best accuracy, sensitivity, and F1 score. We also found that the model achieved comparable performance with about half of the descriptors chosen by feature selection algorithms. We believe that this will contribute to faster and more accurate cocrystal development.

With increasing amount of data, the threat of malware keeps growing recently. The malicious actions embedded in nonexecutable documents especially (e.g., PDF files) can be more dangerous, because it is difficult to detect and most users are not aware of such type of malicious attacks. In this paper, we design a convolutional neural network to tackle the malware detection on the PDF files. We collect malicious and benign PDF files and manually label the byte sequences within the files. We intensively examine the structure of the input data and illustrate how we design the proposed network based on the characteristics of data. The proposed network is designed to interpret high-level patterns among collectable spatial clues, thereby predicting whether the given byte sequence has malicious actions or not. By experimental results, we demonstrate that the proposed network outperform several representative machine-learning models as well as other networks with different settings.

Anesthesia induction is associated with frequent blood pressure fluctuation such as hypotension and hypertension. If it is possible to precisely predict blood pressure a few minutes ahead, anesthesiologists can proactively give anesthetic management before patients develop hemodynamic problem. The objective of this study is to develop a real-time model for predicting 3-min-ahead blood pressure from the start of anesthesia induction to surgical incision. We used only vital signs and anesthesia-related data obtained during anesthesia-induction phase and designed a bidirectional recurrent neural network followed by fully connected layers. We conducted experiments on our collected data of 102 patients, and obtained mean absolute errors between 8.2 mmHg and 11.1 mmHg and standard deviation between 8.7 mmHg and 12.7 mmHg. The average elapsed time for prediction of a batch of 100 unseen data was about 26.56 milliseconds. We believe that this study shows feasibility of real-time prediction of future blood pressures, and the performance will be improved by collecting more data and finding better model structures.

While the exchange of data files or programs on the Internet grows exponentially, most users are vulnerable to infected files, especially to malicious non-executables. Due to the circumstances between South and North Korea, many malicious actions have recently been found in Hangul Word Processor (HWP) non-executable files because the HWP is widely used in schools, military facilities, and government institutions of South Korea. The HWP file usually has one or more byte streams that are often used for the malicious actions. Based on an assumption that infected byte streams have particular patterns, we design a convolutional neural network (CNN) to grasp such patterns. We conduct experiments on our prepared 534 HWP files, and demonstrate that the proposed CNN achieves the best performance compared to other machine learning models. As new malicious attacks keep emerging, we will keep collecting such HWP files and investigate better model structures.

Existing studies in skeleton‐based action recognition mainly utilise skeletal data taken from a single camera. Since the quality of skeletal tracking of a single camera is noisy and unreliable, however, combining data from multiple cameras can improve the tracking quality and hence increase the recognition accuracy. In this study, the authors propose a method called weighted averaging fusion which merges skeletal data of two or more camera views. The method first evaluates the reliability of a set of corresponding joints based on their distances to the centroid, then computes the weighted average of selected joints, that is, each joint is weighted by the overall reliability of the camera reporting the joint. Such obtained, fused skeletal data are used as the input to the action recognition step. Experiments using various frame‐level features and testing schemes show that more than 10% improvement can be achieved in the action recognition accuracy using these fused skeletal data as compared with the single‐view case.

No abstract

Scene text detection is the task of detecting word boxes in given images. The accuracy of text detection has been greatly elevated using deep learning models, especially convolutional neural networks. Previous studies commonly aimed at developing more accurate models, but their models became computationally heavy and worse in efficiency. In this paper, we propose a new efficient model for text detection. The proposed model, namely Compact and Accurate Scene Text detector (CAST), consists of MobileNetV2 as a backbone and balanced decoder. Unlike previous studies that used standard convolutional layers as a decoder, we carefully design a balanced decoder. Through experiments with three well-known datasets, we then demonstrated that the balanced decoder and the proposed CAST are efficient and effective. The CAST was about 1.1x worse in terms of the F1 score, but 30∼115x better in terms of floating-point operations per second (FLOPS).

No abstract