Working Memory Connections for LSTM

Prediction of air pollutant concentrations based on the long short-term memory neural network

In recent years, environmental problems caused by air pollutants have received increasing attention. Effective prediction of air pollutant concentrations is an important way to protect the public from harm. Recently, due to extreme climate and social development, the forest fire frequency has increased. During the biomass combustion process caused by forest fires, the content of particulate matter (PM) in the atmosphere increases significantly. However, most existing air pollutant concentration prediction methods do not consider the considerable impact of forest fires, and effective long-term prediction models have not been established to provide early warnings for harmful gases. Therefore, in this paper, we collected a daily air quality data set (aerodynamic diameter smaller than 2.5 µm, PM2.5) for Heilongjiang Province, China, from 2017 to 2023 and A novel Long Short-Term Memory (LSTM) model was proposed to effectively predict the situation of air pollutants. The model could automatically extract information of the effective time step from the historical data set and combine forest fire disturbance and climate data as auxiliary data to improve the model prediction ability. Moreover, we created artificial neural network (ANN) and permissive regression (support vector machine, SVR) models for comparative experiments. The results showed that the precision accuracy of the developed LSTM model is higher. Unlike the other models, the LSTM neural network model could effectively predict the concentration of air pollutants in long-term series. Regarding long-term observation missions (7 days), the proposed model performed well and stably, with R2 reaching over 88%.

Machine learning in TCM with natural products and molecules: current status and future perspectives

Traditional Chinese medicine (TCM) has been practiced for thousands of years with clinical efficacy. Natural products and their effective agents such as artemisinin and paclitaxel have saved millions of lives worldwide. Artificial intelligence is being increasingly deployed in TCM. By summarizing the principles and processes of deep learning and traditional machine learning algorithms, analyzing the application of machine learning in TCM, reviewing the results of previous studies, this study proposed a promising future perspective based on the combination of machine learning, TCM theory, chemical compositions of natural products, and computational simulations based on molecules and chemical compositions. In the first place, machine learning will be utilized in the effective chemical components of natural products to target the pathological molecules of the disease which could achieve the purpose of screening the natural products on the basis of the pathological mechanisms they target. In this approach, computational simulations will be used for processing the data for effective chemical components, generating datasets for analyzing features. In the next step, machine learning will be used to analyze the datasets on the basis of TCM theories such as the superposition of syndrome elements. Finally, interdisciplinary natural product-syndrome research will be established by unifying the results of the two steps outlined above, potentially realizing an intelligent artificial intelligence diagnosis and treatment model based on the effective chemical components of natural products under the guidance of TCM theory. This perspective outlines an innovative application of machine learning in the clinical practice of TCM based on the investigation of chemical molecules under the guidance of TCM theory.

LCM-Captioner: A lightweight text-based image captioning method with collaborative mechanism between vision and text

Text-based image captioning (TextCap) aims to remedy the shortcomings of existing image captioning tasks that ignore text content when describing images. Instead, it requires models to recognize and describe images from both visual and textual content to achieve a deeper level of comprehension of the images. However, existing methods tend to use numerous complex network architectures to improve performance, which still fails to adequately model the relationship between vision and text on the one side, while on the other side this leads to long running times, high memory consumption, and other unfavorable deployment problems. To solve the above issues, we have developed a lightweight captioning method with a collaborative mechanism, LCM-Captioner, which balances high efficiency with high performance. First, we propose a feature-lightening transformation for the TextCap task, named TextLighT, which is able to learn rich multimodal representations while mapping features to lower dimensions, thereby reducing memory costs. Next, we present a collaborative attention module for visual and text information, VTCAM, to facilitate the semantic alignment of multimodal information to uncover important visual objects and textual content. Finally, the conducted extensive experiments on the TextCaps dataset demonstrate the effectiveness of our method. Code is available at https://github.com/DengHY258/LCM-Captioner.

A Deep Learning Approach to Detect Anomalies in an Electric Power Steering System

As anomaly detection for electrical power steering (EPS) systems has been centralized using model- and knowledge-based approaches, EPS system have become complex and more sophisticated, thereby requiring enhanced reliability and safety. Since most current detection methods rely on prior knowledge, it is difficult to identify new or previously unknown anomalies. In this paper, we propose a deep learning approach that consists of a two-stage process using an autoencoder and long short-term memory (LSTM) to detect anomalies in EPS sensor data. First, we train our model on EPS data by employing an autoencoder to extract features and compress them into a latent representation. The compressed features are fed into the LSTM network to capture any correlated dependencies between features, which are then reconstructed as output. An anomaly score is used to detect anomalies based on the reconstruction loss of the output. The effectiveness of our proposed approach is demonstrated by collecting sample data from an experiment using an EPS test jig. The comparison results indicate that our proposed model performs better in detecting anomalies, with an accuracy of 0.99 and a higher area under the receiver operating characteristic curve than other methods providing a valuable tool for anomaly detection in EPS.

The Scalable Fuzzy Inference-Based Ensemble Method for Sentiment Analysis

Internet environments such as social networks, news sites, and blogs are the platforms where people can share their ideas and opinions. Many people share their comments instantly on the internet, which results in creating large volumes of entries. It is important for institutions and organizations to analyze this big data in an efficient and rapid manner to produce summary information about the feelings or opinions of individuals. In this study, we propose a scalable framework that makes sentiment classification by evaluating the compound probability scores of the most widely used methods in sentiment analysis through a fuzzy inference mechanism in an ensemble manner. The designed fuzzy inference system makes the sentiment estimation by evaluating the compound scores of valance aware dictionary, word embedding, and count vectorization processes. The difference of the proposed method from the classical ensemble methods is that it allows weighting of base learners and combines the strengths of each algorithm through fuzzy rules. The sentiment estimation process from text data can be managed either as a 2-class (positive and negative) or as a 3-class (positive, neutral, and negative) problem. We performed the experimental work on four available tagged social network data sets for both 2-class and 3-class classifications and observed that the proposed method provides improvements in accuracy.

Novel extreme regression-voting classifier to predict death risk in vaccinated people using VAERS data

COVID-19 vaccination raised serious concerns among the public and people are mind stuck by various rumors regarding the resulting illness, adverse reactions, and death. Such rumors are dangerous to the campaign against the COVID-19 and should be dealt with accordingly and timely. One prospective solution is to use machine learning-based models to predict the death risk for vaccinated people and clarify people’s perceptions regarding death risk. This study focuses on the prediction of the death risks associated with vaccinated people followed by a second dose for two reasons; first to build consensus among people to get the vaccines; second, to reduce the fear regarding vaccines. Given that, this study utilizes the COVID-19 VAERS dataset that records adverse events after COVID-19 vaccination as ‘recovered’, ‘not recovered’, and ‘survived’. To obtain better prediction results, a novel voting classifier extreme regression-voting classifier (ER-VC) is introduced. ER-VC ensembles extra tree classifier and logistic regression using soft voting criterion. To avoid model overfitting and get better results, two data balancing techniques synthetic minority oversampling (SMOTE) and adaptive synthetic sampling (ADASYN) have been applied. Moreover, three feature extraction techniques term frequency-inverse document frequency (TF-IDF), bag of words (BoW), and global vectors (GloVe) have been used for comparison. Both machine learning and deep learning models are deployed for experiments. Results obtained from extensive experiments reveal that the proposed model in combination with TF-TDF has shown robust results with a 0.85 accuracy when trained on the SMOTE-balanced dataset. In line with this, validation of the proposed voting classifier on binary classification shows state-of-the-art results with a 0.98 accuracy. Results show that machine learning models can predict the death risk with high accuracy and can assist the authors in taking timely measures.

An air quality index prediction model based on CNN-ILSTM

Air quality index (AQI) is an essential measure of air pollution evaluation, which describes the air pollution degree and its impact on health, so the accurate prediction of AQI is significant. This paper presents an AQI prediction model based on Convolution Neural Networks (CNN) and Improved Long Short-Term Memory (ILSTM), named CNN-ILSTM. ILSTM deletes the output gate in LSTM and improves its input gate and forget gate, and introduces a Conversion Information Module (CIM) to prevent supersaturation in the learning process. ILSTM realizes efficient learning of historical data, improves prediction accuracy, and reduces the training time. CNN extracts the eigenvalues of input data effectively. This paper uses air quality data from 00:00 on January 1, 2017, to 23:00 on June 30, 2021, in Shijiazhuang City, Hebei Province, China, as experimental data sets, and compares this model with eight prediction models: SVR, RFR, MLP, LSTM, GRU, ILSTM, CNN-LSTM, and CNN-GRU to prove the validity and accuracy of CNN-ILSTM prediction model. The experimental results show the MAE of CNN-ILSTM is 8.4134, MSE is 202.1923, R² is 0.9601, and the training time is 85.3 s. In this experiment, the performance of this model performs better than other models.

Deep Learning Application for Effective Classification of Different Types of Psoriasis

Psoriasis is a chronic inflammatory skin disorder mediated by the immune response that affects a large number of people. According to latest worldwide statistics, 125 million individuals are suffering from psoriasis. Deep learning techniques have demonstrated success in the prediction of skin diseases and can also lead to the classification of different types of psoriasis. Hence, we propose a deep learning-based application for effective classification of five types of psoriasis namely, plaque, guttate, inverse, pustular, and erythrodermic as well as the prediction of normal skin. We used 172 images of normal skin from the BFL NTU dataset and 301 images of psoriasis from the Dermnet dataset. The input sample images underwent image preprocessing including data augmentation, enhancement, and segmentation which was followed by color, texture, and shape feature extraction. Two deep learning algorithms of convolutional neural network (CNN) and long short-term memory (LSTM) were applied with the classification models being trained with 80% of the images. The reported accuracies of CNN and LSTM are 84.2% and 72.3%, respectively. A paired sample T-test exhibited significant differences between the accuracies generated by the two deep learning algorithms with a $p<0.001$ . The accuracies reported from this study demonstrate potential of this deep learning application to be applied to other areas of dermatology for better prediction.

Predicting death risk analysis in fully vaccinated people using novel extreme regression-voting classifier

Vaccination for the COVID-19 pandemic has raised serious concerns among the public and various rumours are spread regarding the resulting illness, adverse reactions, and death. Such rumours can damage the campaign against the COVID-19 and should be dealt with accordingly. One prospective solution is to use machine learning-based models to predict the death risk for vaccinated people by utilizing the available data. This study focuses on the prognosis of three significant events including ‘not survived’, ‘recovered’, and ‘not recovered’ based on the adverse events followed by the second dose of the COVID-19 vaccine. Extensive experiments are performed to analyse the efficacy of the proposed Extreme Regression- Voting Classifier model in comparison with machine learning models with Term Frequency-Inverse Document Frequency, Bag of Words, and Global Vectors, and deep learning models like Convolutional Neural Network, Long Short Term Memory, and Bidirectional Long Short Term Memory. Experiments are carried out on the original, as well as, a balanced dataset using Synthetic Minority Oversampling Approach. Results reveal that the proposed voting classifier in combination with TF-IDF outperforms with a 0.85 accuracy score on the SMOTE-balanced dataset. In line with this, the validation of the proposed voting classifier on binary classification shows state-of-the-art results with a 0.98 accuracy.

Explaining transformer-based image captioning models: An empirical analysis

Image Captioning is the task of translating an input image into a textual description. As such, it connects Vision and Language in a generative fashion, with applications that range from multi-modal search engines to help visually impaired people. Although recent years have witnessed an increase in accuracy in such models, this has also brought increasing complexity and challenges in interpretability and visualization. In this work, we focus on Transformer-based image captioning models and provide qualitative and quantitative tools to increase interpretability and assess the grounding and temporal alignment capabilities of such models. Firstly, we employ attribution methods to visualize what the model concentrates on in the input image, at each step of the generation. Further, we propose metrics to evaluate the temporal alignment between model predictions and attribution scores, which allows measuring the grounding capabilities of the model and spot hallucination flaws. Experiments are conducted on three different Transformer-based architectures, employing both traditional and Vision Transformer-based visual features.

Anomaly Detection of the Brake Operating Unit on Metro Vehicles Using a One-Class LSTM Autoencoder

Detecting anomalies in the Brake Operating Unit (BOU) braking system of metro trains is very important for trains’ reliability and safety. However, current periodic maintenance and inspection cannot detect anomalies at an early stage. In addition, constructing a stable and accurate anomaly detection system is a very challenging task. Hence, in this work, we propose a method for detecting anomalies of BOU on metro vehicles using a one-class long short-term memory (LSTM) autoencoder. First, we extracted brake cylinder (BC) pressure data from the BOU data since one of the anomaly cases of metro trains is that BC pressure relief time is delayed by 4 s. After that, extracted BC pressure data is split into subsequences which are fed into our proposed one-class LSTM autoencoder which consists of two LSTM blocks (encoder and decoder). The one-class LSTM autoencoder is trained using training data which only consists of normal subsequences. To detect anomalies from test data that contain abnormal subsequences, the mean absolute error (MAE) for each subsequence is calculated. When the error is larger than a predefined threshold which was set to the maximum value of MAE in the training (normal) dataset, we can declare that example an anomaly. We conducted the experiments with the BOU data of metro trains in Korea. Experimental results show that our proposed method can detect anomalies of the BOU data well.