Sensor-Based Indoor Fire Forecasting Using Transformer Encoder.

Indoor fires may cause casualties and property damage, so it is important to develop a system that predicts fires in advance. There have been studies to predict potential fires using sensor values, and they mostly exploited machine learning models or recurrent neural networks. In this paper, we propose a stack of Transformer encoders for fire prediction using multiple sensors. Our model takes the time-series values collected from the sensors as input, and predicts the potential fire based on the sequential patterns underlying the time-series data. We compared our model with traditional machine learning models and recurrent neural networks on two datasets. For a simple dataset, we found that the machine learning models are better than ours, whereas our model gave better performance for a complex dataset. This implies that our model has a greater potential for real-world applications that probably have complex patterns and scenarios.

Prediction of Postoperative Complications for Patients of End Stage Renal Disease.

End stage renal disease (ESRD) is the last stage of chronic kidney disease that requires dialysis or a kidney transplant to survive. Many studies reported a higher risk of mortality in ESRD patients compared with patients without ESRD. In this paper, we develop a model to predict postoperative complications, major cardiac event, for patients who underwent any type of surgery. We compare several widely-used machine learning models through experiments with our collected data yellow of size 3220, and achieved F1 score of 0.797 with the random forest model. Based on experimental results, we found that features related to operation (e.g., anesthesia time, operation time, crystal, and colloid) have the biggest impact on model performance, and also found the best combination of features. We believe that this study will allow physicians to provide more appropriate therapy to the ESRD patients by providing information on potential postoperative complications.

Malware Detection of Hangul Word Processor Files Using Spatial Pyramid Average Pooling.

Malware detection of non-executables has recently been drawing much attention because ordinary users are vulnerable to such malware. Hangul Word Processor (HWP) is software for editing non-executable text files and is widely used in South Korea. New malware for HWP files continues to appear because of the circumstances between South Korea and North Korea. There have been various studies to solve this problem, but most of them are limited because they require a large amount of effort to define features based on expert knowledge. In this study, we designed a convolutional neural network to detect malware within HWP files. Our proposed model takes a raw byte stream as input and predicts whether it contains malicious actions or not. To incorporate highly variable lengths of HWP byte streams, we propose a new padding method and a spatial pyramid average pooling layer. We experimentally demonstrate that our model is not only effective, but also efficient.

Comparative Analysis on Machine Learning and Deep Learning to Predict Post-Induction Hypotension.

Hypotensive events in the initial stage of anesthesia can cause serious complications in the patients after surgery, which could be fatal. In this study, we intended to predict hypotension after tracheal intubation using machine learning and deep learning techniques after intubation one minute in advance. Meta learning models, such as random forest, extreme gradient boosting (Xgboost), and deep learning models, especially the convolutional neural network (CNN) model and the deep neural network (DNN), were trained to predict hypotension occurring between tracheal intubation and incision, using data from four minutes to one minute before tracheal intubation. Vital records and electronic health records (EHR) for 282 of 319 patients who underwent laparoscopic cholecystectomy from October 2018 to July 2019 were collected. Among the 282 patients, 151 developed post-induction hypotension. Our experiments had two scenarios: using raw vital records and feature engineering on vital records. The experiments on raw data showed that CNN had the best accuracy of 72.63%, followed by random forest (70.32%) and Xgboost (64.6%). The experiments on feature engineering showed that random forest combined with feature selection had the best accuracy of 74.89%, while CNN had a lower accuracy of 68.95% than that of the experiment on raw data. Our study is an extension of previous studies to detect hypotension before intubation with a one-minute advance. To improve accuracy, we built a model using state-of-art algorithms. We found that CNN had a good performance, but that random forest had a better performance when combined with feature selection. In addition, we found that the examination period (data period) is also important.

Investigating the Impact of Possession-Way of a Smartphone on Action Recognition.

For the past few decades, action recognition has been attracting many researchers due to its wide use in a variety of applications. Especially with the increasing number of smartphone users, many studies have been conducted using sensors within a smartphone. However, a lot of these studies assume that the users carry the device in specific ways such as by hand, in a pocket, in a bag, etc. This paper investigates the impact of providing an action recognition system with the information of the possession-way of a smartphone, and vice versa. The experimental dataset consists of five possession-ways (hand, backpack, upper-pocket, lower-pocket, and shoulder-bag) and two actions (walking and running) gathered by seven users separately. Various machine learning models including recurrent neural network architectures are employed to explore the relationship between the action recognition and the possession-way recognition. The experimental results show that the assumption of possession-ways of smartphones do affect the performance of action recognition, and vice versa. The results also reveal that a good performance is achieved when both actions and possession-ways are recognized simultaneously.

Graph Neural Networks with Multi-features for Predicting Cocrystals using APIs and Coformers Interactions.

INTRODUCTION: Active pharmaceutical ingredients (APIs) have gained direct pharmaceutical interest, along with their in vitro properties, and thus utilized as auxiliary solid dosage forms upon FDA guidance and approval on pharmaceutical cocrystals when reacting with coformers, as a potential and attractive route for drug substance development. METHODS: However, screening and selecting suitable and appropriate coformers that may potentially react with APIs to successfully form cocrystals is a time-consuming, inefficient, economically expensive, and labour-intensive task. In this study, we implemented GNNs to predict the formation of cocrystals using our introduced API-coformers relational graph data. We further compared our work with previous studies that implemented descriptor-based models (e.g., random forest, support vector machine, extreme gradient boosting, and artificial neural networks). RESULTS: All built graph-based models show compelling performance accuracies (i.e., 91.36, 94.60 and 95. 95% for GCN, GraphSAGE, and RGCN respectively). RGCN demonstrated effectiveness and prevailed among the built graph-based models due to its capability to capture intricate and learn nuanced relationships between entities such as non-ionic and non-covalent interactions or link information between APIs and coformers which are crucial for accurate predictions and representations. CONCLUSION: These capabilities allows the model to adeptly learn the topological structure inherent in the graph data.

Multi-task learning approach for utilizing temporal relations in natural language understanding tasks.

Various studies have been conducted on multi-task learning techniques in natural language understanding (NLU), which build a model capable of processing multiple tasks and providing generalized performance. Most documents written in natural languages contain time-related information. It is essential to recognize such information accurately and utilize it to understand the context and overall content of a document while performing NLU tasks. In this study, we propose a multi-task learning technique that includes a temporal relation extraction task in the training process of NLU tasks such that the trained model can utilize temporal context information from the input sentences. To utilize the characteristics of multi-task learning, an additional task that extracts temporal relations from given sentences was designed, and the multi-task model was configured to learn in combination with the existing NLU tasks on Korean and English datasets. Performance differences were analyzed by combining NLU tasks to extract temporal relations. The accuracy of the single task for temporal relation extraction is 57.8 and 45.1 for Korean and English, respectively, and improves up to 64.2 and 48.7 when combined with other NLU tasks. The experimental results confirm that extracting temporal relations can improve its performance when combined with other NLU tasks in multi-task learning, compared to dealing with it individually. Also, because of the differences in linguistic characteristics between Korean and English, there are different task combinations that positively affect extracting the temporal relations.

File-level malware detection using byte streams.

As more documents appear on the Internet, it becomes important to detect malware within the documents. Malware of non-executables might be more dangerous because people usually open them without worrying about inherent danger. Recently, deep learning models are used to analyze byte streams of the non-executables for malware detection. Although they have shown successful results, they are commonly designed for stream-level detection, but not for file-level detection. In this paper, we propose a new method that aggregates the stream-level results to get file-level results for malware detection. We demonstrate its effectiveness by experimental results with our annotated dataset, and show that it gives performance gain of 3.37-5.89% of F1 scores.

Novel Cocrystals of Vonoprazan: Machine Learning-Assisted Discovery.

Vonoprazan (VPZ) is the first-in-class potassium-competitive acid blocker (P-CAB), and has many advantages over proton pump inhibitors (PPIs). It is administered as a fumarate salt for the treatment of acid-related diseases, including reflux esophagitis, gastric ulcer, and duodenal ulcer, and for eradication of Helicobacter pylori. To discover novel cocrystals of VPZ, we adopted an artificial neural network (ANN)-based machine learning model as a virtual screening tool that can guide selection of the most promising coformers for VPZ cocrystals. Experimental screening by liquid-assisted grinding (LAG) confirmed that 8 of 19 coformers selected by the ANN model were likely to create new solid forms with VPZ. Structurally similar benzenediols and benzenetriols, i.e., catechol (CAT), resorcinol (RES), hydroquinone (HYQ), and pyrogallol (GAL), were used as coformers to obtain phase pure cocrystals with VPZ by reaction crystallization. We successfully prepared and characterized three novel cocrystals: VPZ-RES, VPZ-CAT, and VPZ-GAL. VPZ-RES had the highest solubility among the novel cocrystals studied here, and was even more soluble than the commercially available fumarate salt of VPZ in solution at pH 6.8. In addition, novel VPZ cocrystals had superior stability in aqueous media than VPZ fumarates, demonstrating their potential for improved pharmaceutical performance.

Prediction of blood pressure changes associated with abdominal pressure changes during robotic laparoscopic low abdominal surgery using deep learning.

BACKGROUND: Intraoperative hypertension and blood pressure (BP) fluctuation are known to be associated with negative patient outcomes. During robotic lower abdominal surgery, the patient's abdominal cavity is filled with CO2, and the patient's head is steeply positioned toward the floor (Trendelenburg position). Pneumoperitoneum and the Trendelenburg position together with physiological alterations during anesthesia, interfere with predicting BP changes. Recently, deep learning using recurrent neural networks (RNN) was shown to be effective in predicting intraoperative BP. A model for predicting BP rise was designed using RNN under special scenarios during robotic laparoscopic surgery and its accuracy was tested. METHODS: Databases that included adult patients (over 19 years old) undergoing low abdominal da Vinci robotic surgery (ovarian cystectomy, hysterectomy, myomectomy, prostatectomy, and salpingo-oophorectomy) at Soonchunhyang University Bucheon Hospital from October 2018 to March 2021 were used. An RNN-based model was designed using Python3 language with the PyTorch packages. The model was trained to predict whether hypertension (20% increase in the mean BP from baseline) would develop within 10 minutes after pneumoperitoneum. RESULTS: Eight distinct datasets were generated and the predictive power was compared. The macro-average F1 scores of the datasets ranged from 68.18% to 72.33%. It took only 3.472 milliseconds to obtain 39 prediction outputs. CONCLUSIONS: A prediction model using the RNN may predict BP rises during robotic laparoscopic surgery.

Antimalarial Drug Predictions Using Molecular Descriptors and Machine Learning against Plasmodium Falciparum.

Malaria remains by far one of the most threatening and dangerous illnesses caused by the plasmodium falciparum parasite. Chloroquine (CQ) and first-line artemisinin-based combination treatment (ACT) have long been the drug of choice for the treatment and controlling of malaria; however, the emergence of CQ-resistant and artemisinin resistance parasites is now present in most areas where malaria is endemic. In this work, we developed five machine learning models to predict antimalarial bioactivities of a drug against plasmodium falciparum from the features (i.e., molecular descriptors values) obtained from PaDEL software from SMILES of compounds and compare the machine learning models by experiments with our collected data of 4794 instances. As a consequence, we found that three models amongst the five, namely artificial neural network (ANN), extreme gradient boost (XGB), and random forest (RF), outperform the others in terms of accuracy while observing that, using roughly a quarter of the promising descriptors picked by the feature selection algorithm, the five models achieved equivalent and comparable performance. Nevertheless, the contribution of all molecular descriptors in the models was investigated through the comparison of their rank values by the feature selection algorithm and found that the most potent and relevant descriptors which come from the 'Autocorrelation' module contributed more while the 'Atom type electrotopological state' contributed the least to the model.

Machine-Learning-Based Approach to Differential Diagnosis in Tuberculous and Viral Meningitis.

BACKGROUND: Tuberculous meningitis (TBM) is the most severe form of tuberculosis, but differentiating between the diagnosis of TBM and viral meningitis (VM) is difficult. Thus, we have developed machine-learning modules for differentiating TBM from VM. MATERIAL AND METHODS: For the training data, confirmed or probable TBM and confirmed VM cases were retrospectively collected from five teaching hospitals in Korea between January 2000 - July 2018. Various machine-learning algorithms were used for training. The machine-learning algorithms were tested by the leave-one-out cross-validation. Four residents and two infectious disease specialists were tested using the summarized medical information. RESULTS: The training study comprised data from 60 patients with confirmed or probable TBM and 143 patients with confirmed VM. Older age, longer symptom duration before the visit, lower serum sodium, lower cerebrospinal fluid (CSF) glucose, higher CSF protein, and CSF adenosine deaminase were found in the TBM patients. Among the various machine-learning algorithms, the area under the curve (AUC) of the receiver operating characteristics of artificial neural network (ANN) with ImperativeImputer for matrix completion (0.85; 95% confidence interval 0.79 - 0.89) was found to be the highest. The AUC of the ANN model was statistically higher than those of all the residents (range 0.67 - 0.72, P <0.001) and an infectious disease specialist (AUC 0.76; P = 0.03). CONCLUSION: The machine-learning techniques may play a role in differentiating between TBM and VM. Specifically, the ANN model seems to have better diagnostic performance than the non-expert clinician.