Term Memory Research Articles

Probing intrinsic noise correlation time in non-Markovian diffusive systems.

The generalized Langevin equation describes molecular motion in complex systems using memory and random noise terms. Memory effects, the inherent time correlation in random noise, significantly influence molecular diffusive behaviors. However, estimating the intrinsic noise correlation time remains challenging because of difficulties in measuring the memory term. We propose a metric to probe the noise correlation time based on the deviation between characteristic times of configurational diffusion and "diffusion motion" in the kinetic energy space. This approach stems from the observation that memory effects delay relaxation time between displacement and velocity response functions. Our metric relies solely on the velocity autocorrelation function, commonly used in experimental model parameterization. Both analytical and numerical results for various physical models demonstrate its effectiveness in probing noise correlation time. Furthermore, we apply this metric to study complex diffusive phenomena, including non-exponential relaxation in molecular hydrodynamics and anomalous diffusion in crowded environments. By comparing with system's relaxation time, we reveal that long-range noise correlations play a key role in these non-trivial diffusive phenomena.

Optimized LSTM Model for Accurate Blood Glucose Prediction in Type-1 Diabetes

Objective: Blood glucose levels in diabetes are influenced by various factors, making prediction challenging when considering many variables. Therefore, this study proposes an optimized LSTM model to improve the prediction accuracy of blood glucose using a single variable (CGM), thereby simplifying diabetes management. Methods: This study has implemented the LSTM model to predict blood glucose levels in type-1 diabetics for a 60-minute prediction horizon. The grid search technique has been implemented to fine-tune the hyperparameters of the LSTM model. Ohio Datasets were used for this study. Root mean squared error (RMSE) and error grid analysis (EGA) were used for the evaluation of model performance. Findings: The proposed study reveals 26.13 ± 3.25 mg/dl predicting accuracy for a 60-minute prediction horizon. Clark Error grid analysis (CEGA) was used to validate clinical acceptance of the proposed model, and the results provided higher than 97% prediction accuracy in the Ohio dataset. Novelty: The patient-specific optimized model has outperformed the conventional LSTM model. The results show a significant improvement in model performance compared to previous work in terms of root mean square error (RMSE). Keywords: Long-short term memory (LSTM), Grid search optimization technique, Continuous Glucose monitors (CGM), Blood Glucose, Type-1 Diabetes

Twitter Bot Detection Using Machine Learning and Deep Learning Techniques

Abstract—The proliferation of Twitter bots poses a serious threat to the reliability of online conversations and results in disinformation, spam, and opinion manipulation. This paper presents a comprehensive examination of Twitter bot detection techniques with traditional machine learning (ML) algorithms contrasted with cutting-edge deep learning (DL) models. Key fea- tures like tweet frequency, follower-following ratios, user behavior patterns, and content features are investigated. We compare algorithms like Random Forest, Support Vector Machines (SVM), Logistic Regression, K-Nearest Neighbors (KNN), Long Short- Term Memory (LSTM), and Recurrent Neural Networks (RNN) based on accuracy, precision, recall, and F1-score metrics. Our experiments showed that Random Forest was the best with the highest accuracy and thus, it is the best-suited model for the dataset used in this experiment. We also address the issues of real- time bot detection, the limitation of single models, and suggest a hybrid approach that takes advantage of the strengths of both ML and DL approaches for better performance. Index Terms—Twitter Bot Detection, Machine Learning, Deep Learning, Social Network Analysis, Random Forest, Support Vec- tor Machine (SVM), Logistic Regression, K-Nearest Neighbors (KNN), Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNN).

Resolving Data Imbalance Using a Bi-Directional Long-Short Term Memory for Enhanced Diabetes Mellitus Detection

Diabetes is the body’s inability to efficiently break down sugar or secrete enough insulin required to process glucose, which supports normal bodily functions. Diabetes, as a prevalent chronic disorder, has contributed to numerous underlying health challenges among its carriers and is classified by the WHO as the world’s deadliest disease and silent killer. Its non-communicable nature makes early diagnosis difficult, allowing progression through various stages: type I, type II, pre-diabetes, and gestational. This challenge is further compounded by the imbalanced nature of diabetes datasets, which leads to high misclassification, poor generalization, and reduced accuracy. This study predicts diabetes using a bi-directional long short-term memory (BiLSTM) model applied to two datasets: (a) PIMA Indian Diabetes and (b) Iraqi Society Dataset, to evaluate the impact of six known balancing techniques and assess their effectiveness. Results show that for PID, the SMOTE-Tomek fused BiLSTM outperforms other balancing schemes with F1, Accuracy, Precision, Recall, and Specificity scores of 0.9182, 0.9198, 0.9128, 0.9248, and 0.9208, respectively. For ISD, it also achieves the best performance with values of 0.9367, 0.9369, 0.9386, 0.9388, and 0.9313, respectively. Other balancing approaches yielded F1 scores ranging from [0.6751 to 0.9347], accuracy [0.684 to 0.9358], Precision [0.6851 to 0.9296], Recall [0.6639 to 0.9356], and specificity [0.6658 to 0.9298]. These results imply that BiLSTM is resilient to the vanishing gradient problem and can effectively classify diabetes cases with enhanced performance.

Establishing a periodic SM model with Fourier analysis for enhancing global soil moisture forecasting

Accurately predicting global soil moisture (SM) is crucial for sustainable agriculture and water resource management. Recognizing the challenges posed by the heterogeneity of SM’s spatiotemporal variability, this study proposes a novel approach that leverages Fourier analysis to decompose the periodic fluctuations in SM, revealing underlying trends and cycles. This approach is integrated with Long Short Term Memory (LSTM) networks to enhance the accuracy of global SM prediction. Fourier analysis transforms time series data of SM into frequencies and amplitudes, capturing its intrinsic periodic characteristics. This transformation reveals both variable and invariant features representing changes within and between periods. By integrating these periodic features with sequence data and leveraging the memory and sequence learning capabilities of LSTM neural networks, the accuracy and reliability of global SM prediction can be enhanced. Our experiments on the LandBench1.0 dataset demonstrate that the proposed model reduces the root mean square error by to compared to the state-of-the-art methods. This study underscores that the LSTM with periodic features of SM can adapt to the inherent complex spatial-temporal patterns in SM dynamics, especially in scenarios characterized by rapid environmental changes and subtle temporal dynamics.

An attention based hybrid approach using CNN and BiLSTM for improved skin lesion classification

Skin lesions remain a significant global health issue, with their incidence rising steadily over the past few years. Early and accurate detection is crucial for effective treatment and improving patient outcomes. This work explores the integration of advanced Convolutional Neural Networks (CNNs) with Bidirectional Long Short Term Memory (BiLSTM) enhanced by spatial, channel, and temporal attention mechanisms to improve the classification of skin lesions. The hybrid model is trained to distinguish between various skin lesions with high precision. Among the models evaluated, the CNN (original architecture) with BiLSTM and attention mechanisms model achieved the highest performance, with an accuracy of 92.73%, precision of 92.84%, F1 score of 92.70%, recall of 92.73%, Jaccard Index (JAC) of 87.08%, Dice Coefficient (DIC) of 92.70%, and Matthews Correlation Coefficient (MCC) of 91.55%. The proposed model was compared to other configurations, including CNN with Gated Recurrent Units (GRU) and attention mechanisms, CNN with LSTM and attention mechanisms, CNN with BiGRU and attention mechanisms, CNN with BiLSTM, CNN with LSTM, CNN with BiGRU, CNN with GRU, standalone CNN, InceptionV3, Visual Geometry Group-16 (VGG16), and Xception, to highlight the efficacy of the proposed approach. This research aims to empower healthcare professionals by providing a robust diagnostic tool that enhances accuracy and supports proactive management strategies. The model’s ability to analyze high-resolution images and capture complex features of skin lesions promises significant advancements in early detection and personalized treatment. This work not only seeks to advance the technological capabilities in skin lesion diagnostics but also aims to mitigate the disease’s impact through timely interventions and improved healthcare outcomes, ultimately enhancing public health resilience on a global scale.

Prediction of super-large diameter shield attitude based on LSTM-Transformer

Accurate control of shield tunneling attitude is a critical technology for ensuring construction safety and tunnel quality. With the rapid development of urban underground space, the construction of ultra-large-diameter shield tunnels has become increasingly common. However, the high precision requirements for attitude control and the complex dynamic responses during construction pose significant challenges, which traditional prediction methods struggle to address. To tackle this technical challenge, this study proposes an LSTM-Transformer prediction model that integrates long short term memory (LSTM) networks for temporal feature extraction with Transformer’s global attention mechanism. The model predicts four key shield attitude parameters and is validated using field data from the Jiangyin-Jingjiang Yangtze river tunnel project, with comparative analysis against existing models. The research results show that: (1) The LSTM-Transformer attitude prediction model achieves an R2 value of 0.881 and a mean absolute error (MAE) value of 2.24 mm, outperforming existing models in prediction accuracy; (2) Feature importance analysis reveals the key parameters that should be prioritized during shield attitude adjustment, providing a theoretical basis for dynamic attitude control; (3) The model effectively provides early warnings for shield attitude deviation risks, significantly enhancing construction safety and efficiency. The research provides important theoretical support and practical references for shield attitude prediction and risk warning in similar engineering projects.

An efficient patient’s response predicting system using multi-scale dilated ensemble network framework with optimization strategy

The forecasting of a patient’s response to radiotherapy and the likelihood of experiencing harmful long-term health impacts would considerably enhance individual treatment plans. Due to the continuous exposure to radiation, cardiovascular disease and pulmonary fibrosis might occur. For forecasting the response of patients to chemotherapy, the Convolutional Neural Networks (CNN) technique is widely used. With the help of radiotherapy, cancer diseases are diagnosed, but some patients suffer from side effects. The toxicity of radiotherapy and chemotherapy should be estimated. For validating the patient’s improvement in treatments, a patient response prediction system is essential. In this paper, a Deep Learning (DL) based patient response prediction system is developed to effectively predict the response of patients, predict prognosis and inform the treatment plans in the early stage. The necessary data for the response prediction are collected manually. The collected data are then processed through the feature selection segment. The Repeated Exploration and Exploitation-based Coati Optimization Algorithm (REE-COA) is employed to select the features. The selected weight features are input into the prediction process. Here, the prediction is performed by Multi-scale Dilated Ensemble Network (MDEN), where we integrated Long-Short term Memory (LSTM), Recurrent Neural Network (RNN) and One-dimensional Convolutional Neural Networks (1DCNN). The final prediction scores are averaged to develop an effective MDEN-based model to predict the patient’s response. The proposed MDEN-based patient’s response prediction scheme is 0.79%, 2.98%, 2.21% and 1.40% finer than RAN, RNN, LSTM and 1DCNN, respectively. Hence, the proposed system minimizes error rates and enhances accuracy using a weight optimization technique.

Sequential recommendation via agent-based irrelevancy skipping.

Sequential recommendation via agent-based irrelevancy skipping.

Implementasi Sistem Prediksi Saham Real-Time dengan Integrasi Yahoo Finance API dan Machine Learning di Google Colab

The issue of stock price prediction is a critical topic in the financial world, where accurate predictions are essential for making better investment decisions. This research develops a real-time stock price prediction system by integrating the Yahoo Finance API and machine learning algorithms, executed on the Google Colab platform. This system allows for direct retrieval of stock market data from Yahoo Finance to analyze data patterns and generate more accurate stock price predictions. The methods used include collecting historical and real-time data from the Yahoo Finance API, data preprocessing, training models using Long Short Term Memory (LSTM), validating the model with K-Fold Cross Validation, and evaluating performance using various standard metrics. The development and implementation of the model are carried out on Google Colab. The results show that the LSTM model can provide high-accuracy stock price predictions. This system significantly contributes to improving stock price prediction accuracy and helps investors make better decisions based on real-time data

An attention driven long short term memory based multi-attribute feature learning for shot boundary detection

An attention driven long short term memory based multi-attribute feature learning for shot boundary detection

Joint prediction of multi-aircraft trajectories in terminal airspace: A Flight Pattern-Guided Social Long-Short Term Memory network

Joint prediction of multi-aircraft trajectories in terminal airspace: A Flight Pattern-Guided Social Long-Short Term Memory network

Emotional stimulated speech-based assisted early diagnosis of depressive disorders using personality-enhanced deep learning.

Emotional stimulated speech-based assisted early diagnosis of depressive disorders using personality-enhanced deep learning.

A Mathematical Model of COVID-19 Using Piecewise Derivative of Fractional Order

Currently the dynamical systems of infectious disease were studied by using various definitions of fractional calculus. Because the mentioned area has the ability to demonstrate the short and long memory terms involve in the physical dynamics of numerous real world problems. In this work, we consider a seven compartmental model for the transmission dynamics of COVID-19 including susceptible (S ), vaccinated (V), exposed(E), infected (I), quarantined (Q), recovered (R), and death (D) classes. We first revisit the fundamental outcomes related to equilibrium points, basic reproduction numbers, sensitivity analysis and equilibrium points including disease free (DF), and endemic equilibrium points (EE). In addition, our main goal to investigate the consider model under the new aspect of fractional calculus known as piecewise fractional order operators. The mentioned operators have the ability to demonstrate the multi phase behaviours of the dynamical problems. The said characteristics cannot be described by using the traditional operators. We apply the tools of nonlinear analysis to deduce sufficient conditions for the existence of at least one solution and its uniqueness. Additionally, we also investigate the results related to stability analysis of Ulam-Hyers (U-H)type. Finally, we extend the concepts of RK2 method to form a sophisticated algorithms to simulate the results graphically. We present the numerical results for various fractional order.

Hexatalk using ANN and DNNS

Speaker recognition is an essential aspect of human-computer interaction, with applications in security, personalized services, and more. This project proposes an end-to-end speaker recognition system leveraging Long Short- Term Memory (LSTM) neural networks. Mel-Frequency Cepstral Coefficients (MFCCs) are used as audio features, processed by an LSTM model to classify speakers with high accuracy. The proposed system demonstrates the efficacy of LSTM for temporal feature analysis, achieving robust performance in noisy environments.

Temporal Gap Filling of Nighttime Light Composites

The temporal nighttime light (NTL) data generated by DMSP-OLS sensors was discovered to have large gaps (missing values) over time. The research aims to provide a scientifically valid gap-filling mechanism for having consistent DMPS-OLS time series data (1992-2013) and predicting the historic NTL (1991-1985) for long-term studies. A deep learning neural network, Long Short Term Memory (LSTM) has been proposed in the study for temporal gap filling and historic NTL prediction. The developed LSTM model is being tested in a time distributed wrapper way having window size (3-7) for the temporal gap filling and prediction of the historic NTL. According to the accuracy evaluation, the developed model has a testing accuracy of R2 = 0.96 with a window size of 5. The historic population, Gross Domestic Product (GDP), and Electric Power Consumption per capita (EPC) data are utilized to validate the gap-filled and historic NTL. R2 = 0.91 w.r.t population, R2 = 0.71 w.r.t GDP, and R2 = 0.69 w.r.t EPC, is been found during the assessment of these parameters with the sum of light of the year (1985-2013). The historic & gap-filled NTL data can be used in various studies to monitor temporal development.

Energy Consumption Prediction for Smart Homes

Abstract - Accurate electricity consumption forecasting is a critical component in ensuring the stability, efficiency, and cost- effectiveness of national power systems. In Finland, where the electricity sector is shaped by nuclear energy, cogeneration, renewable sources like black liquor and wood, and significant electricity imports, reliable forecasting becomes even more essential. This research investigates the application of Long Short- Term Memory (LSTM), a type of recurrent neural network (RNN) designed for sequential data, in predicting future energy consumption trends. A univariate time series dataset from Fin- grid, Finland’s transmission system operator, encompassing six years of hourly electricity consumption data, serves as the basis for model training and evaluation. The study aims to assess the model’s performance in capturing temporal patterns and generating accurate short-term forecasts. Our results highlight the capability of LSTM models to effectively model complex energy usage patterns and provide meaningful insights for energy planning and decision-making processes. Key Words: Long short-term memory (LSTM), Time series analysis, Machine learning, Energy prediction, Fingrid, Univariate data, Finland electricity sector.

Exploring the dynamics of Shannon's information and iconicity in language processing and lexeme evolution.

This two-part meta-study explores the relationship of Shannon's information and iconicity in American English, with a focus on their implications for cognitive processing and the evolution of lexemes. Part one explores the expression of information in iconic words by calculating phonemic bigram surprisal using a very large corpus of spoken American English and cross referencing it with iconicity ratings. Iconic words-those with a form/meaning resemblance-are known to be processed with a cognitive advantage, so they are included in our tests as a benchmark. Within the framework of the Iconic Treadmill Hypothesis, we posit that as iconic words evolve towards arbitrariness, bigram sequences become more predictable, offsetting some the cognitive costs associated with processing arbitrary words. In part 2, we extend Cognitive Load Theory and the Lossy Context Surprisal Model-both sentence level language processing models-to test our predictions at the bigram level using the results of a battery of existing psycholinguistic experiments. In line with these models that explain the psycholinguistic consequences of hearing improbable words in sentences, our results show that words made up of improbable phonemes are processed with cognitive disadvantage, but that extra processing effort enhances their retention in long term memory. Overall, our findings speak to the cognitive limitations of language processing and how these limitations influence lexeme evolution.

High accuracy indoor positioning system using Galois field-based cryptography and hybrid deep learning

In smart manufacturing, logistics, and other inside settings where the Global Positioning System (GPS) doesn’t work, indoor positioning systems (IPS) are essential. Due to environmental complexity, signal noise, and possible data manipulation, traditional IPS techniques struggle with accuracy, resilience, and security. Online and offline phases are distinguished in the suggested indoor location system that employs deep learning and fingerprinting. During the offline phase, mobile devices gather signal strength measurements and contextual data traverse inside settings via Wi-Fi, Bluetooth, and magnetometers. Fingerprint classification using Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering follows the application of signal processing techniques for noise reduction and data augmentation. The online phase involves extracting information to improve the model’s accuracy. These features can be signal-based, spatial–temporal, motion-based, or environmental. The Deep Spatial–Temporal Attention Network (Deep-STAN) is an innovative hybrid model for location classification that combines Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), Long-Short Term Memory (LSTMs), and attention processes. The model hyperparameters are fine-tuned using hybrid optimization to guarantee optimal performance. The work’s main contribution is the incorporation of ECC, an effective encryption and decryption method for signal data, which is based on Galois fields. This cryptographic method is well-suited for real-world applications since it guarantees low-latency operations while simultaneously improving data integrity and confidentiality. In addition, S-box enhances the IPS’s resilience and security by including QR codes for distinct location marking and blockchain technology for safe and immutable storing of positioning data. Moreover, the performance of the suggested model includes an accuracy of 0.9937, precision of 0.987, sensitivity of 0.9898, and specificity of 0.9878, while when 80% of data were used it had an accuracy of 0.9804, precision of 0.9722, sensitivity of 0.9859, and specificity of 0.9756. These outcomes prove that the proposed system is stable and flexible enough to be used in indoor positioning applications.

Analysis of Fine Food Reviews Using Intelligent Computing Model

Automated overall review score based on text descriptions. This would provide companies a quick and accurate way to measure customer satisfaction without needing manual inspection and analysis. We used the LSTM (Long Short Term Memory), RNN (Recurrent Neural Network), and GRU (Gated Recurrent Unit) designs, three widely used recurrent neural network (RNN) architectures, to do this. Tasks requiring sequence modeling, such as natural language processing, are ideally suited for these systems. The "Amazon Fine Food Reviews" dataset that we acquired from Kaggle was first preprocessed. In order to do this, the dataset needed to be cleaned up by having the comments' special characters and punctuation removed. To create a fair and understandable dataset, we additionally chose a subset of the data depending on the duration of the reviews. In addition, we used word clouds for exploratory data analysis to understand the distribution of the most common terms. Next, we utilized the preprocessed dataset to train our LSTM, RNN, and GRU models. Based on the input text descriptions, these models were trained to predict the total review score. In order to reduce the loss value and increase accuracy, the model parameters were optimized throughout the training phase. We evaluated the effectiveness of our models utilizing the testing group. The findings demonstrated that our model produced effectively a loss value of 0.2 for the testing group. This suggests that our algorithm can pretty accurately and effectively estimate the overall review score based on the text descriptions.This approach may be applied practically to automate the generation of an overall evaluation score in the food sector. The model may provide an impartial evaluation of customer happiness by examining the text descriptions that consumers have submitted. Based on user input, it can assist companies in monitoring and enhancing the quality of their goods or services. Overall, our study shows that LSTM, RNN, and GRU models are capable of accurately predicting review scores based on text descriptions. The model's accuracy and loss values suggest that it may be useful in automating review analysis and measuring customer happiness.