Dynamic Neural Network Research Articles

Optimizing Deep Learning-Based Natural Language Processing for Sentiment Analysis

Deep learning (DL)-based natural language processing (NLP) technologies are applied and optimized for the analysis of emotional implications in text data. Understanding emotional intricacies in text has become necessary for a variety of industries, like social media analysis, customer feedback interpretation, and mental health monitoring, due to the exponential rise of digital material. The study emphasizes the difficulties associated with the automation of emotion analysis, including the diversities of the text formats and the NLP complexities. This research gathers that text-based emotion analysis data and pre-processing techniques are utilized for the analysis of emotion, including normalizing methods such as tokenization and lemmatization, along with the removal of stop words and punctuation. The model incorporates pre-trained embeddings and attention mechanisms to capture contextual information and emotional subtleties for any language. The study proposes the hybrid model using tuned crow search optimized dynamic graph neural network (TCSO-DGNN) on an automatically optimized process to enhance the accuracy of emotion detection. The proposed model results on emotion analysis data and superior results are obtained, indicating better performance in precision (94.72%), accuracy (96.43%), recall (94.36%), and F1-score (93.02%). The findings indicate that the inclusion of TCSO-DGNN significantly enhances the identification of complex emotional expressions concerning traditional NLP techniques. This study explores the potential of integrated DL approaches to refine the capabilities of emotion analysis tools, particularly in contexts requiring refined interpretations of human emotions expressed online. Additionally, it contributes to academic discourse in the field of emotion analysis while providing practical insights for developers of NLP applications across business and public service sectors.

SAT-GATv2: A Dynamic Attention-Based Graph Neural Network for Solving Boolean Satisfiability Problem

We propose SAT-GATv2, a graph neural network (GNN)-based model designed to solve the Boolean satisfiability problem (SAT) through graph-based deep learning techniques. SAT-GATv2 transforms SAT formulas into graph structures, leveraging message-passing neural networks (MPNNs) to propagate local information and dynamic attention mechanisms (GATv2) to accurately capture inter-node dependencies and enhance node feature representations. Unlike traditional heuristic-driven SAT solvers, SAT-GATv2 adopts a data-driven approach, learning structural patterns directly from graph representations and providing a complementary framework to existing methods. Experimental results demonstrate that SAT-GATv2 achieves an accuracy improvement of 1.75–5.51% over NeuroSAT on challenging random 3-SAT(n) instances, highlighting its effectiveness in handling difficult problem distributions, and outperforms other GNN-based models on SR(n) datasets, showcasing its scalability and adaptability. Ablation studies validate the critical roles of MPNNs and GATv2 in improving prediction accuracy and scalability. While SAT-GATv2 does not yet surpass CDCL-based solvers in overall performance, it addresses their limitations in scalability and adaptability to complex instances, offering an efficient graph-based alternative for tackling larger and more complex SAT problems. This study establishes a foundation for integrating deep learning with combinatorial optimization, emphasizing its potential for applications in artificial intelligence and operations research.

Spatiotemporal Interaction Based Dynamic Adversarial Adaptive Graph Neural Network for Air-Quality Prediction

Air quality issues have become a major environmental concern, with severe air pollution significantly reducing air quality and posing threats to human health. Accurate air quality prediction is crucial for preventing individuals from suffering the detrimental effects of severe air pollution. Recently, deep learning methods based on spatiotemporal graph neural networks (GNNs) have made considerable progress in modeling the temporal and spatial dependencies within air quality data by integrating GNNs with sequential models. Unfortunately, previous work often treats temporal and spatial dependencies as independent components, neglecting the intricate interactions between them. This oversight prevents the models from fully exploiting the complex spatiotemporal dependencies in the data, adversely affecting their predictive performance. To address these issues, we propose a general spatiotemporal interaction framework for air quality prediction. This framework models the bidirectional interactions between temporal and spatial dependencies in a data-driven manner. Furthermore, we designed a spatiotemporal feature extraction module and a dynamic adversarial adaptive graph learning module based on this framework. We introduce the Spatial-Temporal Interaction based Dynamic Adversarial Adaptive Graph Neural Network, capable of capturing the complex interactions between spatiotemporal dependencies and learning the dynamic spatial topology among sites by incorporating the competitive optimization concept of generative adversarial networks. Extensive experiments on two real-world datasets demonstrate the effectiveness of the proposed method, outperforming existing baseline models.

ANALYZING SOCIAL MEDIA SENTIMENT TOWARD SPECIFIC COMMODITIES FOR FORECASTING PRICE MOVEMENTS IN COMMODITY MARKETS

This study adopts a systematic literature review to analyze social media sentiment towards specific commodities to enhance the accuracy of price movement forecasts in commodity markets. Drawing from the field of applied mathematics, the research gathered literature from Scopus, DOAJ, and Google Scholar databases, covering publications from 2014 to 2024. A rigorous search strategy yielded 66 journal articles, with 30 being selected for their close relevance to keywords such as "social media sentiment," "commodity markets," and "price forecasting." Results indicate that social media sentiment significantly influences commodity prices, with particular variations based on commodity type and geographical context. Specific sentiment factors—especially intensity, polarity, and timing—were found to have a pronounced impact on price dynamics, with sentiment polarity being particularly influential in volatile markets. Additionally, advanced analytical methods, like Bayesian Dynamic Linear Models and LSTM neural networks, enhance predictive accuracy when applied to sentiment analysis in this context. These findings underscore the value of social media sentiment in refining forecasting models, while also highlighting gaps in understanding regional sentiment variations and their effects on different commodity types. By synthesizing these insights, this study emphasizes the importance of considering social media sentiment for more accurate price predictions and identifies key areas for future research to explore the multifaceted impacts of sentiment in commodity markets.

Effect of Inter Turbine Spacing in Omnidirectional Wind for Savonius Cluster: A Computational Fluid Dynamics and Artificial Neural Network Approach

Effect of Inter Turbine Spacing in Omnidirectional Wind for Savonius Cluster: A Computational Fluid Dynamics and Artificial Neural Network Approach

Benchmarking the speed-accuracy tradeoff in object recognition by humans and neural networks.

Active object recognition, fundamental to tasks like reading and driving, relies on the ability to make time-sensitive decisions. People exhibit a flexible tradeoff between speed and accuracy, a crucial human skill. However, current computational models struggle to incorporate time. To address this gap, we present the first dataset (with 148 observers) exploring the speed-accuracy tradeoff (SAT) in ImageNet object recognition. Participants performed a 16-way ImageNet categorization task where their responses counted only if they occurred near the time of a fixed-delay beep. Each block of trials allowed one reaction time. As expected, human accuracy increases with reaction time. We compare human performance with that of dynamic neural networks that adapt their computation to the available inference time. Time is a scarce resource for human object recognition, and finding an appropriate analog in neural networks is challenging. Networks can repeat operations by using layers, recurrent cycles, or early exits. We use the repetition count as a network's analog for time. In our analysis, the number of layers, recurrent cycles, and early exits correlates strongly with floating-point operations, making them suitable time analogs. Comparing networks and humans on SAT-fit error, category-wise correlation, and SAT-curve steepness, we find cascaded dynamic neural networks most promising in modeling human speed and accuracy. Surprisingly, convolutional recurrent networks, typically favored in human object recognition modeling, perform the worst on our benchmark.

Method for predicting remaining useful life of rolling bearings based on dynamic complexity characteristic entropy and quantum neural networks

Method for predicting remaining useful life of rolling bearings based on dynamic complexity characteristic entropy and quantum neural networks

A dynamic hypergraph attention network: Capturing market-wide spatiotemporal dependencies for stock selection

As more companies choose to go public to raise additional funds, investors are confronted with the challenge of choosing high-quality stocks from a multitude of options to generate returns on investment. In recent years, various deep learning models have been utilized to explore spatiotemporal dependencies through the analysis of historical price indicators of individual stocks and related stocks for stock selection. However, despite advancements in these models, they neglect the dynamic influence of global market-wide information on spatiotemporal dependencies of stocks. Focusing solely on stock-specific dependencies limits their ability to learn comprehensive spatiotemporal representations. To address this limitation, we propose a novel market-wide dynamic hypergraph attention network (MDHAN) for stock selection. The key of MDHAN is introducing an attention-based market-impacted GRU (att-MGRU) and an attention-based market-impacted dynamic hypergraph neural network (att-MHGNN). Specifically, The att-MGRU is a variant of GRU, which employs a fusion gate to capture daily market-wide temporal influences on each stock and a temporal attention mechanism to discern the contributions of historical influences across different trading days. The att-MHGNN first constructs a dynamic market-impacted hypergraph to model the market’s influence on stock relationships, which is then fed into an attention-based hypergraph neural network to extract the spatial dependencies influenced by the market. By conducting experiments on four real world datasets from China and the US, covering a span of four years, we demonstrate that MDHAN significantly outperforms state-of-the-art methods.

Novel Speech-Based Emotion Climate Recognition in Peers’ Conversations Incorporating Affect Dynamics and Temporal Convolutional Neural Networks

Novel Speech-Based Emotion Climate Recognition in Peers’ Conversations Incorporating Affect Dynamics and Temporal Convolutional Neural Networks

Towards sludge bulking diagnosis via dynamic attention graph neural network

Towards sludge bulking diagnosis via dynamic attention graph neural network

Exploring Computer-Aided Environmental Art Design: A Course Overview

Computer aided education is transforming with the integration of technology. In the context of advancing art education, there is a pressing need for innovation to enhance student engagement and learning outcomes. This study introduces an innovative approach by employing an Adaptive Kookaburra Optimized Dynamic Recurrent Neural Network (AKO-DRNN) with the framework of computer-aided environmental art design courses. The traditional methods of teaching art are being complemented by computer-aided tools and intelligent systems. This research explores the application of AKO-DRNN in revolutionizing art education, focusing on environmental art design. The primary goal is to develop an instructional system that leverages advanced algorithms to offer personalized, accurate aesthetic guidance, enhance creative exploration, and elevate students' practical skills in environmental art design. This study integrates AKO-DRNN into the course structure, which combines deep learning (DL) models with environmental design principles. The AKO-DRNN model utilizes dynamic recurrent networks optimized by a Kookaburra-inspired optimization algorithm to effectively analyze and predict artistic styles and features. This model provides real-time feedback and adaptive learning paths tailored to individual student needs. Implementation of the suggested model has demonstrated significant improvements in students’ design quality, creativity, and skill acquisition. The adaptive nature of the model enhances learning outcomes and engagement. The established framework offers a robust solution for modernizing art education in environmental design, fostering greater innovation and practical skills among students.

Forecasting and managing urban EV charging demand with spatiotemporal graph models

The rapid proliferation of electric vehicles (EVs) necessitates advanced infrastructural adaptations to accommodate increased demand. This paper introduces a novel dynamic spatiotemporal graph neural network (D-ST-GNN) model designed to optimize the placement and management of EV charging stations in urban environments. Leveraging real-time data and graph theory, the D-ST-GNN model dynamically represents the evolving urban EV charging landscape, incorporating factors such as traffic patterns, station connectivity, and user demand. Employing a synthesized dataset that simulates the interactions of 1000 charging stations across a metropolitan area, our model integrates diverse data sources including traffic flow, weather conditions, and user interactions. The model processes these data through advanced machine learning techniques, utilizing graph neural networks (GNNs) to capture the complex dependencies within the network. The predictive capability of the model was rigorously tested over a 30-day period, demonstrating a mean absolute error of 7.5 kW and a root mean square error of 10 kW, affirming its high accuracy and reliability in forecasting daily charging demands. Significantly, the model employs real-time adaptive algorithms that recalibrate the network based on observed data, enhancing the predictive accuracy over time. Through simulation, the model has shown potential to reduce queue times at charging stations by up to 30% during peak hours and increase overall network efficiency by 25%. The practical application of the D-ST-GNN model provides urban planners and policymakers with a robust tool for making informed decisions regarding the expansion and optimization of EV charging infrastructures.

Used economy market insight: Sailboat industry pricing mechanism and regional effects.

With the popularity of circular economy around the world, transactions in the second-hand sailboat market are extremely active. Determining pricing strategies and exploring their regional effects is a blank area of existing research and has important practical and statistical significance. Therefore, this article uses the random forest model and XGBoost algorithm to identify core price indicators, and uses an innovative rolling NAR dynamic neural network model to simulate and predict second-hand sailboat price data. On this basis, we also constructed a regional effect multi-level model (RE-MLM) from three levels: geography, economy and country to clarify the impact of geographical areas on sailboat prices. The research results show that, first of all, the price of second-hand sailboats fluctuates greatly, and the predicted value better reflects the overall average price level. Secondly, there are significant regional differences in price levels across regions, economies and ethnic groups. Therefore, the price of second-hand sailboats is affected by many factors and has obvious regional effects. In addition, the model evaluation results show that the model constructed in this study has good accuracy, validity, portability and versatility, and can be extended to price simulation and regional analysis of different markets in different regions.

Bayesian approaches for revealing complex neural network dynamics in Parkinson’s disease

Bayesian approaches for revealing complex neural network dynamics in Parkinson’s disease

SLC35A2 loss of function variants affect glycomic signatures, neuronal fate, and network dynamics.

SLC35A2 encodes a UDP-galactose transporter essential for glycosylation of proteins and galactosylation of lipids and glycosaminoglycans. Germline genetic SLC35A2 variants have been identified in congenital disorders of glycosylation and somatic SLC35A2 variants have been linked to intractable epilepsy associated with malformations of cortical development. However, the functional consequences of these pathogenic variants on brain development and network integrity remain elusive. In this study, we use an isogenic human induced pluripotent stem cell-derived neuron model to comprehensively interrogate the functional impact of loss of function variants in SLC35A2 through the integration of cellular and molecular biology, protein glycosylation analysis, neural network dynamics, and single cell electrophysiology. We show that loss of function variants in SLC35A2 result in disrupted glycomic signatures and precocious neurodevelopment, yielding hypoactive, asynchronous neural networks. This aberrant network activity is attributed to an inhibitory/excitatory imbalance as characterization of neural composition revealed preferential differentiation of SLC35A2 loss of function variants towards the GABAergic fate. Additionally, electrophysiological recordings of synaptic activity reveal a shift in excitatory/inhibitory balance towards increased inhibitory drive, indicating changes occurring specifically at the pre-synaptic terminal. Our study is the first to provide mechanistic insight regarding the early development and functional connectivity of SLC35A2 loss of function variant harboring human neurons, providing important groundwork for future exploration of potential therapeutic interventions.

Dynamic Neural Network States During Social and Non-Social Cueing in Virtual Reality Working Memory Tasks: A Leading Eigenvector Dynamics Analysis Approach.

This research investigates brain connectivity patterns in reaction to social and non-social stimuli within a virtual reality environment, emphasizing their impact on cognitive functions, specifically working memory. Employing the LEiDA framework with EEG data from 47 participants, I examined dynamic brain network states elicited by social avatars compared to non-social stick cues during a VR memory task. Through the integration of LEiDA with deep learning and graph theory analyses, unique connectivity patterns associated with cue type were discerned, underscoring the substantial influence of social cues on cognitive processes. LEiDA, conventionally utilized with fMRI, was creatively employed in EEG to detect swift alterations in brain network states, offering insights into cognitive processing dynamics. The findings indicate distinct neural states for social and non-social cues; notably, social cues correlated with a unique brain state characterized by increased connectivity within self-referential and memory-processing networks, implying greater cognitive engagement. Moreover, deep learning attained approximately 99% accuracy in differentiating cue contexts, highlighting the efficacy of prominent eigenvectors from LEiDA in EEG analysis. Analysis of graph theory also uncovered structural network disparities, signifying enhanced integration in contexts involving social cues. This multi-method approach elucidates the dynamic influence of social cues on brain connectivity and cognition, establishing a basis for VR-based cognitive rehabilitation and immersive learning, wherein social signals may significantly enhance cognitive function.

Dynamic Convolutional Neural Networks Based on Adaptive 2D Memristors

AbstractConvolutional Neural Networks (CNNs) are pivotal in modern digital computing, particularly for tasks like image classification, inspired by the receptive fields of the human brain. Nevertheless, CNNs implemented on conventional digital computers face significant limitations due to inflexible kernels that cannot adjust to dynamic inputs, and the von Neumann architecture, which leads to inefficient data transfer between memory and processing units. This research presents a hardware‐software co‐designed solution, a Dynamic Convolutional Neural Network (dCNN), empowered by three‐terminal adaptive two‐dimensional (2D) memristors. These memristors consist of a vertical heterostructure integrating silver, an atomically thin insulator (CrPS4), and graphene as a semimetal. This configuration allows for the dynamic tuning of conductive filament properties, emulating the heterosynaptic plasticity observed in biological neural systems. The three‐terminal memristor design permits the dCNN to actively adjust kernel weights in its attention layer according to the input stimuli. The empirical tests demonstrate that image classification accuracy using our adaptive 2D memristor‐enhanced dVGG reaches up to 94% on the CIFAR‐10 dataset, which exceeds the performance of static VGG. Furthermore, the energy efficiency of our dVGG significantly outperforms that of GPUs, aligning more closely with the energy dynamics of the human brain in terms of both consumption and classification accuracy.

Multiscale Cloud-Based Pipeline for Neuronal Electrophysiology Analysis and Visualization.

Electrophysiology offers a high-resolution method for real-time measurement of neural activity. Longitudinal recordings from high-density microelectrode arrays (HD-MEAs) can be of considerable size for local storage and of substantial complexity for extracting neural features and network dynamics. Analysis is often demanding due to the need for multiple software tools with different runtime dependencies. To address these challenges, we developed an open-source cloud-based pipeline to store, analyze, and visualize neuronal electrophysiology recordings from HD-MEAs. This pipeline is dependency agnostic by utilizing cloud storage, cloud computing resources, and an Internet of Things messaging protocol. We containerized the services and algorithms to serve as scalable and flexible building blocks within the pipeline. In this paper, we applied this pipeline on two types of cultures, cortical organoids and ex vivo brain slice recordings to show that this pipeline simplifies the data analysis process and facilitates understanding neuronal activity.

Chaotic Dynamics of a Two-Neuron Memristive Hopfield Neural Network under Electromagnetic Radiation

Electromagnetic radiation (EMR) is a recognized approach for investigating the behavior of the nervous system. This research explores a Hopfield Neural Network (HNN) with two neurons, examining the interplay between synapses and hyperbolic memristors, along with the effects of EMR. By modifying the interference among the networks and adjusting synaptic weights, we can control neuronal capabilities. The study simulates synaptic interference between the two neurons, incorporating parameters of weight and memory, and analyzes how EMR influences chaotic dynamics, complex behavior, transient disturbances, phase portraits, chaotic phenomena, and branching diagrams within these neural networks. This paper investigates how electromagnetic radiation (EMR) influences chaotic dynamics in a two-neuron-based memristive Hopfield neural network (HNN) with synaptic crosstalk. The dynamic behavior of the HNN can be regulated by altering the EMR input to the affected neuron. The proposed model has been simulated using PSpice. The findings demonstrate that external stimuli, represented by EMR, can both enhance complex dynamic behaviors and suppress chaotic patterns by adjusting parameters. Finally, circuit experiments using PSpice confirm the feasibility of the theoretical model, contributing to the control of chaotic phenomena.

Multivariate time series classification based on spatial-temporal attention dynamic graph neural network

Multivariate time series classification based on spatial-temporal attention dynamic graph neural network