Advanced Neural Network Architectures Research Articles

Stock Market Prediction

The prediction of stock market trends is a challenging yet critical task in the financial sector, given its significant implications for investors, traders, and financial institutions. This research leverages the Long Short-Term Memory (LSTM) algorithm, a type of recurrent neural network (RNN), to develop a robust model for forecasting stock prices. The study utilizes historical stock market data sourced from Yahoo Finance, accessed via the yfinance package in Python. The primary objectives are to preprocess the data, implement the LSTM model, and evaluate its performance against traditional models such as Random Forest and Linear Regression. Data preprocessing involved handling missing values, normalizing the dataset, and transforming it into sequences suitable for LSTM training. The model's architecture includes multiple LSTM layers designed to capture temporal dependencies in the data. The study evaluates the model's performance using metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and prediction accuracy. Comparative analysis shows that the LSTM model outperforms both Random Forest and Linear Regression models, with lower MSE and RMSE values and higher accuracy in predicting stock prices. This research discovered that LSTM's ability to retain long-term dependencies makes it particularly effective for stock market prediction, where historical trends and patterns significantly influence future prices. The results indicate that the LSTM model provides more reliable and precise predictions, which can enhance decision-making in trading and investment. This research highlights the potential of advanced neural network architectures in financial forecasting, offering a valuable tool for investors aiming to optimize their strategies and mitigate risks. The significance of this study lies in its practical application in the financial industry, demonstrating that machine learning models, particularly LSTM, can substantially improve the accuracy of stock market predictions. Future research could explore the integration of additional features, such as macroeconomic indicators and sentiment analysis, to further enhance model performance. This study underscores the importance of continuous innovation and the adoption of sophisticated algorithms to navigate the complexities of financial markets.

Classification of Various Plant Leaf Disease Using Pretrained Convolutional Neural Network On Imagenet

Introduction/Background Plant diseases and pernicious insects are a considerable threat in the agriculture sector. Leaf diseases impact agricultural production. Therefore, early detection and diagnosis of these diseases are essential. This issue can be addressed if a farmer can detect the diseases properly. Objective The fundamental goal of this project is to create and test a model for precisely classifying leaf diseases in plants. Materials and Methods This paper introduces a model designed to classify leaf diseases effectively. The research utilizes the publicly available PlantVillage dataset, which includes 38 different classes of leaf images, ranging from healthy to disease-infected leaves. Pretrained CNN (Convolutional Neural Network) models, including VGG16, ResNet50, InceptionV3, MobileNetV2, AlexNet, and EfficientNet, are employed for image classification. Results The paper provides a performance comparison of these models. The results show that the EfficientNet model achieves an accuracy of 97.5% in classifying healthy and diseased leaf images, outperforming other models. Discussion This research highlights the potential of utilizing advanced neural network architectures for accurate disease detection in the agricultural sector. Conclusion This study demonstrates the efficacy of employing sophisticated CNN models, particularly EfficientNet, to properly identify leaf diseases. Such technological developments have the potential to improve disease detection in agriculture. These improvements help to improve food security by allowing for preventive actions to battle crop diseases.

ACCIDENT DETECTION USING TENSOR FLOW AND CNN

Accident detection is a critical aspect of road safety management, with timely identification and response being paramount to minimizing casualties and damage. In this project, a comprehensive approach to accident detection utilizing Deep learning techniques is proposed. The system aims to automatically detect accidents in real time using information obtained from diverse sources like video feeds, sensor inputs, and audio signals. The project involves the development and implementation of Deep learning models trained on labeled datasets containing examples of accident and non-accident scenarios. Supervised learning techniques like classification, regression and advanced neural network architecture like convolution, recurrent neural network are considered to effectively process and analyze the data. The deployed system continuously monitors the environment, processing incoming data streams to identify patterns indicative of accidents. Upon detection, the system triggers appropriate response mechanisms, such as alerting emergency services or nearby vehicles, facilitating swift intervention. The project aims to contribute to road safety initiatives by providing a scalable and efficient solution for accident detection. Additionally, the insights gained from the data collected by the system can inform policymakers and urban planners in implementing measures to prevent accidents and improve overall road safety. Through this project, the students gain valuable experience in applying Deep learning techniques to address real-world challenges and contribute to societal welfare.

Advanced Neural Network Architecture for Enhanced Multi-Lead ECG Arrhythmia Detection through Optimized Feature Extraction

Cardiovascular diseases are a pervasive global health concern, contributing significantly to morbidity and mortality rates worldwide. Among these conditions, arrhythmia, characterized by irregular heart rhythms, presents formidable diagnostic challenges. This study introduces an innovative approach utilizing deep learning techniques, specifically Convolutional Neural Networks (CNNs), to address the complexities of arrhythmia classification. Leveraging multi-lead Electrocardiogram (ECG) data, our CNN model, comprising six layers with a residual block, demonstrates promising outcomes in identifying five distinct heartbeat types: Left Bundle Branch Block (LBBB), Right Bundle Branch Block (RBBB), Atrial Premature Contraction (APC), Premature Ventricular Contraction (PVC), and Normal Beat. Through rigorous experimentation, we highlight the transformative potential of our methodology in enhancing diagnostic accuracy for cardiovascular arrhythmias. Arrhythmia diagnosis remains a critical challenge in cardiovascular care, often relying on manual interpretation of ECG signals, which can be time-consuming and prone to subjectivity. To address these limitations, we propose a novel approach that leverages deep learning algorithms to automate arrhythmia classification. By employing advanced CNN architectures and multi-lead ECG data, our methodology offers a robust solution for precise and efficient arrhythmia detection. Through comprehensive evaluation, we demonstrate the effectiveness of our approach in facilitating more accurate clinical decision-making, thereby improving patient outcomes in managing cardiovascular arrhythmias.

PREDICTING STOCK MARKET MOVEMENTS USING NEURAL NETWORKS: A REVIEW AND APPLICATION STUDY

In the rapidly evolving landscape of financial markets, the quest for accurate stock market predictions has never been more critical. This paper delves into the transformative potential of neural network models in forecasting stock market movements, offering a comprehensive examination of their effectiveness compared to traditional predictive models. With a focus on the evolution of stock market prediction methodologies, this study aims to uncover the nuanced dynamics of neural networks, their comparative analysis with other models, and the pivotal role of data preprocessing in enhancing prediction accuracy. Employing a qualitative analysis framework, the research meticulously synthesizes findings from selected studies, highlighting the superior performance of neural network models in capturing complex market patterns and adapting to volatility. The results underscore the significant impact of data quality and quantity, architectural nuances of neural networks, and the strategic implications for investors navigating the stock market's unpredictability. Despite the promising outcomes, the study acknowledges inherent challenges in the real-world application of these models, including data imperfections and the complexity of financial ecosystems. Conclusively, the paper advocates for ongoing innovation, interdisciplinary collaboration, and the strategic integration of advanced neural network architectures to overcome existing limitations. Recommendations emphasize the critical need for high-quality, diverse datasets and continuous model refinement to harness the full predictive power of neural networks in stock market forecasting. This study not only illuminates the path forward for investors and financial analysts but also sets the stage for future research in this dynamic field.
 Keywords: Neural Network Models, Stock Market Prediction, Data Preprocessing, Model Accuracy, Financial Markets, Innovation.

Enhancement of handwritten text recognition using AI-based hybrid approach

Handwritten text recognition (HTR) within computer vision and image processing stands as a prominent and challenging research domain, holding significant implications for diverse applications. Among these, it finds usefulness in reading bank checks, prescriptions, and deciphering characters on various forms. Optical character recognition (OCR) technology, specifically tailored for handwritten documents, plays a pivotal role in translating characters from a range of file formats, encompassing both word and image documents. Challenges in HTR encompass intricate layout designs, varied handwriting styles, limited datasets, and less accuracy achieved.Recent advancements in Deep Learning and Machine Learning algorithms, coupled with the vast repositories of unprocessed data, have propelled researchers to achieve remarkable progress in HTR. This paper aims to address the challenges in handwritten text recognition by proposing a hybrid approach. The primary objective is to enhance the accuracy of recognizing handwritten text from images. Through the integration of Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) with a Connectionist Temporal Classification (CTC) decoder, the results indicate substantial improvement. The proposed hybrid model achieved an impressive 98.50% and 98.80% accuracy on the IAM and RIMES datasets, respectively. This underscores the potential and efficacy of the consecutive use of these advanced neural network architectures in enhancing handwritten text recognition accuracy.•The proposed method introduces a hybrid approach for handwritten text recognition, employing CNN and BiLSTM with CTC decoder.•Results showcase a remarkable accuracy improvement of 98.50% and 98.80% on IAM and RIMES datasets, emphasizing the potential of this model for enhanced accuracy in recognizing handwritten text from images.

Adaptive compound power quality disturbance detection via OMD and improved networks for renewable energy systems

In the evolving landscape of power systems, the integration of various renewable energy resources (RERs) introduces complex challenges, particularly in maintaining power quality, which are paramount for system stability. To address this issue, an adaptive power quality disturbance (PQD) detection framework is implemented in this paper. First, the optimal mode decomposition (OMD) is developed to decompose the compound PQDs into sub-ingredients to make them more visible based on the optimal energy ratio. Subsequently, we propose an improved attention convolutional neural network (IACNN), an advanced neural network architecture that leverages an enhanced attention mechanism to expedite the identification of PQDs. Importantly, the sub-ingredients can be strengthened based on the established PQD detection framework. Finally, a series of experiments are conducted under different noise levels and various types of PQDs. The results demonstrate that the proposed framework has profound detection effectivity with about 99.2% accuracy under the simulation condition of 20 dB noise level. In addition, the experimental verification analysis proves a satisfactory real-time performance. This underscores the potential of the proposed framework as a significant advancement in the realm of power quality management, offering a robust solution to the challenges posed by the integration of RERs into modern power systems.

Predicting stock market movements using neural networks: A review and application study

In the rapidly evolving landscape of financial markets, the quest for accurate stock market predictions has never been more critical. This paper delves into the transformative potential of neural network models in forecasting stock market movements, offering a comprehensive examination of their effectiveness compared to traditional predictive models. With a focus on the evolution of stock market prediction methodologies, this study aims to uncover the nuanced dynamics of neural networks, their comparative analysis with other models, and the pivotal role of data preprocessing in enhancing prediction accuracy. Employing a qualitative analysis framework, the research meticulously synthesizes findings from selected studies, highlighting the superior performance of neural network models in capturing complex market patterns and adapting to volatility. The results underscore the significant impact of data quality and quantity, architectural nuances of neural networks, and the strategic implications for investors navigating the stock market's unpredictability. Despite the promising outcomes, the study acknowledges inherent challenges in the real-world application of these models, including data imperfections and the complexity of financial ecosystems. Conclusively, the paper advocates for ongoing innovation, interdisciplinary collaboration, and the strategic integration of advanced neural network architectures to overcome existing limitations. Recommendations emphasize the critical need for high-quality, diverse datasets and continuous model refinement to harness the full predictive power of neural networks in stock market forecasting. This study not only illuminates the path forward for investors and financial analysts but also sets the stage for future research in this dynamic field.

Unveiling the Power of Deep Learning: Insights into Advanced Neural n Network Architectures

Deep learning has emerged as a powerful paradigm within artificial intelligence, revolutionizing various fields with its ability to automatically discover intricate patterns and representations from vast amounts of data. In this article, we delve into the latest advancements in deep learning, focusing on the exploration of advanced neural network architectures. Through a comprehensive review and analysis, we uncover the underlying principles and methodologies driving the success of these architectures, elucidating their capabilities and limitations. From convolutional neural networks (CNNs) for image recognition to recurrent neural networks (RNNs) for sequential data processing, we highlight key innovations and breakthroughs that have propelled the field forward. Furthermore, we discuss recent trends such as attention mechanisms, transformer models, and graph neural networks, showcasing their potential applications and impact across diverse domains. By unraveling the power of deep learning and its advanced neural network architectures, this article aims to provide valuable insights into the state-of-the-art techniques driving AI research and development.

Enhancing Electrical Load Prediction Using a Bidirectional LSTM Neural Network

Precise anticipation of electrical demand holds crucial importance for the optimal operation of power systems and the effective management of energy markets within the domain of energy planning. This study builds on previous research focused on the application of artificial neural networks to achieve accurate electrical load forecasting. In this paper, an improved methodology is introduced, centering around bidirectional Long Short-Term Memory (LSTM) neural networks (NN). The primary aim of the proposed bidirectional LSTM network is to enhance predictive performance by capturing intricate temporal patterns and interdependencies within time series data. While conventional feed-forward neural networks are suitable for standalone data points, energy consumption data are characterized by sequential dependencies, necessitating the incorporation of memory-based concepts. The bidirectional LSTM model is designed to furnish the prediction framework with the capacity to assimilate and leverage information from both preceding and forthcoming time steps. This augmentation significantly bolsters predictive capabilities by encapsulating the contextual understanding of the data. Extensive testing of the bidirectional LSTM network is performed using multiple datasets, and the results demonstrate significant improvements in accuracy and predictive capabilities compared to the previous simpleRNN-based framework. The bidirectional LSTM successfully captures underlying patterns and dependencies in electrical load data, achieving superior performance as gauged by metrics such as root mean square error (RMSE) and mean absolute error (MAE). The proposed framework outperforms previous models, achieving a remarkable RMSE, attesting to its remarkable capacity to forecast impending load with precision. This extended study contributes to the field of electrical load prediction by leveraging bidirectional LSTM neural networks to enhance forecasting accuracy. Specifically, the BiLSTM’s MAE of 0.122 demonstrates remarkable accuracy, outperforming the RNN (0.163), LSTM (0.228), and GRU (0.165) by approximately 25%, 46%, and 26%, in the best variation of all networks, at the 24-h time step, while the BiLSTM’s RMSE of 0.022 is notably lower than that of the RNN (0.033), LSTM (0.055), and GRU (0.033), respectively. The findings highlight the significance of incorporating bidirectional memory and advanced neural network architectures for precise energy consumption prediction. The proposed bidirectional LSTM framework has the potential to facilitate more efficient energy planning and market management, supporting decision-making processes in power systems.

Battle Models: Inception ResNet vs. Extreme Inception for Marine Fish Object Detection

Within the domain of deep learning applied to computer vision, there exists a significant emphasis on the competition between two prominent models, namely Inception ResNet and Xception, particularly in the field of marine fish object detection. The present study conducted a comparative analysis of two advanced neural network architectures in order to assess their efficacy in the identification and localization of marine fish species in underwater images. The two models underwent a rigorous evaluation, utilizing their capabilities in feature extraction. The findings indicate a complex performance landscape, wherein Inception ResNet exhibits remarkable accuracy in identifying marine fish objects, while Xception demonstrates superior computational efficiency. The present study elucidates the inherent trade-off between precision and computational expenditure, offering valuable perspectives on the pragmatic ramifications of choosing one model over another. Furthermore, this research underscores the significance of carefully choosing a suitable model that aligns with the particular requirements of object detection applications in the context of marine fish. This study endeavors to guide professionals and scholars in marine biology and computer vision, enabling them to make well-informed choices when utilizing deep learning techniques to detect maritime fish objects in underwater settings. The research specifically focuses on the comparison between Inception ResNet and Xception models.

Consumer complaints of consumer financial protection bureau via two-stage residual one-dimensional convolutional neural network (TSR1DCNN)

The Consumer Financial Protection Bureau (CFPB) is a government body responsible for safeguarding consumers from financial fraud and abuse. Managing customer complaints is one of the key tasks undertaken by the CFPB. However, the sheer volume of complaints received can overwhelm the bureau's resources, hindering prompt and efficient resolution. To address this challenge, we propose a novel approach called the Two-Stage Residual One-Dimensional Convolutional Neural Network (TSR1DCNN) to optimize the processing of consumer complaints at the CFPB. In this study, we conducted comprehensive experiments, including Ablation Experiment 1 (AE1) and Ablation Experiment 2 (AE2), to evaluate the effectiveness of our proposed TSR1DCNN model. AE1 involved removing the first Conv1D layer, while AE2 removed the Batch Normalization layer. These experiments allowed us to assess the impact of removing specific components on the overall performance of the model. Furthermore, we compared our TSR1DCNN model with other popular deep learning architectures, including 1DCNN, LSTM, and BLSTM, to provide a comprehensive analysis of our proposed approach. Using a dataset of 555,957 consumer complaints received by the CFPB, we trained and tested the TSR1DCNN model, as well as the ablated versions in AE1 and AE2, alongside the 1DCNN, LSTM, and BLSTM models. The results showed that the TSR1DCNN model achieved an impressive accuracy of 78.07% on the training set and 76.53% on the test set. In comparison, AE1 achieved an accuracy of 69.63% with a loss of 1.1207, while AE2 achieved an accuracy of 71.00% with a loss of 1.0583. The performance of the TSR1DCNN model outperformed the other deep learning architectures, including 1DCNN, LSTM, and BLSTM, indicating its superiority in handling consumer complaints effectively. These results demonstrate the superiority of the TSR1DCNN model over the ablated versions in AE1 and AE2, as well as its superiority over other commonly used deep learning architectures. By incorporating advanced neural network architectures such as 1DCNN, LSTM, and BLSTM, and considering the specific modules where our proposed method operates, we provide a promising solution for enhancing the efficiency and effectiveness of complaint-handling processes in organizations facing a large volume of complaints, such as the CFPB.

A novel hybrid face mask detection approach using Transformer and convolutional neural network models.

Face and face mask detection are one of the most popular topics in computer vision literature. Face mask detection refers to the detection of people's faces in digital images and determining whether they are wearing a face mask. It can be of great benefit in different domains by ensuring public safety through the monitoring of face masks. Current research details a range of proposed face mask detection models, but most of them are mainly based on convolutional neural network models. These models have some drawbacks, such as their not being robust enough for low quality images and their being unable to capture long-range dependencies. These shortcomings can be overcome using transformer neural networks. Transformer is a type of deep learning that is based on the self-attention mechanism, and its strong capabilities have attracted the attention of computer vision researchers who apply this advanced neural network architecture to visual data as it can handle long-range dependencies between input sequence elements. In this study, we developed an automatic hybrid face mask detection model that is a combination of a transformer neural network and a convolutional neural network models which can be used to detect and determine whether people are wearing face masks. The proposed hybrid model's performance was evaluated and compared to other state-of-the-art face mask detection models, and the experimental results proved the proposed model's ability to achieve a highest average precision of 89.4% with an execution time of 2.8 s. Thus, the proposed hybrid model is fit for a practical, real-time trial and can contribute towards public healthcare in terms of infectious disease control.

A Comparative Study of Speaker Role Identification in Air Traffic Communication Using Deep Learning Approaches

Automatic spoken instruction understanding (SIU) of the controller-pilot conversations in the air traffic control (ATC) requires not only recognizing the words and semantics of the speech but also determining the role of the speaker. However, few of the published works on the automatic understanding systems in air traffic communication focus on speaker role identification (SRI). In this article, we formulate the SRI task of controller-pilot communication as a binary classification problem. Furthermore, the text-based, speech-based, and speech-and-text-based multi-modal methods are proposed to achieve a comprehensive comparison of the SRI task. To ablate the impacts of the comparative approaches, various advanced neural network architectures are applied to optimize the implementation of text-based and speech-based methods. Most importantly, a multi-modal speaker role identification network (MMSRINet) is designed to achieve the SRI task by considering both the speech and textual modality features. To aggregate modality features, the modal fusion module is proposed to fuse and squeeze acoustic and textual representations by modal attention mechanism and self-attention pooling layer, respectively. Finally, the comparative approaches are validated on the ATCSpeech corpus collected from a real-world ATC environment. The experimental results demonstrate that all the comparative approaches worked for the SRI task, and the proposed MMSRINet shows competitive performance and robustness compared with the other methods on both seen and unseen data, achieving 98.56% and 98.08% accuracy, respectively.

Comparative Analysis of Hidden Markov Model and Bidirectional Long Short-Term Memory for POS Tagging in Eastern Armenian

This scientific article introduces a novel POS tagging system specifically developed for the Eastern Armenian language. The primary objective of the study is to conduct a comparative analysis of two well-established methods for part-of-speech (POS) tagging: the hidden Markov model (HMM) coupled with the Viterbi algorithm, and an artificial neural network in the form of a recurrent neural network. The study places particular emphasis on the Eastern Armenian language and employs the ArmSpeech-POS Eastern Armenian part-of-speech tagged corpus for conducting comprehensive experiments and evaluations. POS tagging is a fundamental task in natural language processing (NLP) that involves assigning grammatical tags to words in a given text. Accurate POS tagging is crucial for various NLP applications, including machine translation, information retrieval, and sentiment analysis. The Viterbi algorithm is a well-established probabilistic method that utilizes a hidden Markov model (HMM) to determine the most likely sequence of POS tags. On the other hand, RNNs, a type of deep learning model, can capture complex patterns and dependencies in sequential data. Experimental results indicate that both methods achieve reasonable accuracy in POS tagging for Eastern Armenian. However, the RNN outperforms the Viterbi algorithm, exhibiting higher accuracy rates. This can be attributed to the RNN’s ability to capture long-range dependencies and learn intricate linguistic patterns. The article concludes by discussing the implications of the study’s findings and potential areas for further research. It emphasizes the significance of accurate POS tagging for improving NLP applications in Eastern Armenian and suggests exploring advanced neural network architectures and incorporating linguistic features to enhance POS tagging performance.

Channel-Robust Specific Emitter Identification Based on Transformer

Specific emitter identification (SEI) refers to the process of identifying emitter individuals based on corresponding wireless signals. Although deep learning has been successfully applied in SEI, the performance remains to be improved when the channel changes. In this paper, we suggest that a potential reason of the performance degradation is inadequacy of model capacity. Therefore, Transformer, an advanced neural network architecture with large model capacity, is applied for channel-robust SEI. Experimental results show that Transformer achieves better performance than conventional convolutional neural networks.

A deep learning approach utilizing clinical and molecular data for identifying prognostic biomarkers in patients treated with immune checkpoint inhibitors: An ORIEN pan-cancer study.

2619 Background: Immune checkpoint inhibitors (ICIs) have made significant improvements in the treatment of cancer patients (pts), but many continue to experience primary or secondary resistance. Here, we leveraged clinical and genomic data to identify prognostic biomarkers in pts treated with ICIs utilizing a pan-cancer approach. Methods: Pts were enrolled to the Total Cancer Care protocol across 18 cancer centers within the Oncology Research Information Exchange Network (ORIEN). RNA-seq was performed on tumors following the RSEM pipeline and gene expressions were quantified as Transcript Per Million (TPM) and were logarithmically normalized. An Auto-Encoder Survival Deep Network (AE-SDN) architecture was developed that combined the reconstruction loss of AE with Cox regression for modeling time to event. For comparison, immunoscore for each pt was calculated based on the estimated densities of tumor CD3+ and CD8+ T cells (Galon, 2020) utilizing CIBERSORTx. The quality of overall survival (OS) predictions was assessed using Harrell’s concordance index (C-index). Log-rank test was used to assess stratified group differences (by ICI or cancer histology) along with Kaplan-Meier (KM) survival analysis of AE-SDN and immunoscore. Results: Pts (n=522) with 4 cancer types including melanoma (n=125), renal cell carcinoma (n=149), non-small cell lung cancer (n=128) and head and neck cancer (n=120) treated with 6 ICI regimens were included in this analysis. ICI regimens were nivolumab (n=219), pembrolizumab (n=202), ipilimumab+nivolumab (n=69), ipilimumab (n=30), avelumab (n=1) and cemiplimab (n=1). The Table summarizes the overall C-index and associated 95% CIs and log-rank P values for the entire cohort (regardless of histology) resulting from our proposed AE-SDN model and the separate estimated immunoscore categorization. AE-SDN top selected genes were mostly related to immunity, carcinogenesis and tumor suppression. The corresponding KM plots showed significantly wider separations of the survival curves in favor of our proposed AE-SDN model relative to the immunoscore with more than 20% improvement in prediction power. Conclusions: Deep network machine learning analysis is a promising approach to identifying relevant prognostic biomarkers in cancer pts treated with ICI. This may lead to novel therapeutic predictive signatures and identification of mechanisms of ICI resistance. Our AE-SDN gene expression signature was significantly prognostic and outperformed the estimated CD3+, CD8+ T Cell immunoscore. Further refinements to our prediction power are ongoing along with more advanced neural network architectures to elucidate related functional pathways. [Table: see text]

3D joints estimation of human body using part segmentation

In this paper, we propose a novel method for 3D human joint estimation using part segmentation, and introduce an application for size measurement based on the obtained joints. A human segmentation dataset is first prepared as training set for the advanced neural network architecture. Different human parts yielded from the neural network are utilized to extract human joints. In the proposed method, the joints are categorized into the active joints and inert joints. In the extraction process of the active joints, the mathematical analysis method is adopted to calculate the joint positions. The geometric features of different human segments are further used to extract the inert joints. Moreover, we test on the dataset to compare its performance with our previous method based on geometrical features. The results show the average error of the joints is less than 4.2 cm, which is significantly improved from 5.8 cm demonstrated in our previous research. We also investigate the human size measurement. The distance between the joints is used to calculate the length, and the ellipse fitting method based on multi-frame point cloud is adapted to calculate the human girths. Compared with the manual measurement data, the size error is less than 4.1 cm.

Sequence Alignment Ensemble With a Single Neural Network for Sequence Labeling

Sequence labeling, in which a class or label is assigned to each token in a given input order, is a fundamental task in natural language processing. Many advanced neural network architectures have recently been proposed to solve the sequential labeling problem affecting this task. By contrast, only a few approaches have been proposed to address the sequential ensemble problem. In this paper, we resolve the sequential ensemble problem by applying the sequential alignment method in a proposed ensemble framework. Specifically, we propose a simple but efficient ensemble candidate generation framework with which multiple heterogeneous systems can easily be prepared from a single neural sequence labeling network. To evaluate the proposed framework, experiments were conducted with part-of-speech (POS) tagging and dependency label prediction problems. The results indicate that the proposed framework achieved accuracy values that were higher by 0.19 and 0.33 than those achieved by the hard-voting method on the Penn-treebank POS-tagged and Universal dependency-tagged datasets, respectively.

ThreshNet: An Efficient DenseNet Using Threshold Mechanism to Reduce Connections

With the continuous development of neural networks for computer vision tasks, more and more network architectures have achieved outstanding success. As one of the most advanced neural network architectures, DenseNet shortcuts all feature maps to solve the model depth problem. Although this network architecture has excellent accuracy with low parameters, it requires an excessive inference time. To solve this problem, HarDNet reduces the connections between the feature maps, making the remaining connections resemble harmonic waves. However, this compression method may result in a decrease in the model accuracy and an increase in the parameters and model size. This network architecture may reduce the memory access time, but its overall performance can still be improved. Therefore, we propose a new network architecture, ThreshNet, using a threshold mechanism to further optimize the connection method. Different numbers of connections for different convolution layers are discarded to accelerate the inference of the network. The proposed network has been evaluated with image classification using CIFAR 10 and SVHN datasets under platforms of NVIDIA RTX 3050 and Raspberry Pi 4. The experimental results show that, compared with HarDNet68, GhostNet, MobileNetV2, ShuffleNet, and EfficientNet, the inference time of the proposed ThreshNet79 is 5%, 9%, 10%, 18%, and 20% faster, respectively. The number of parameters of ThreshNet95 is 55% less than that of HarDNet85. The new model compression and model acceleration methods can speed up the inference time, enabling network models to operate on mobile devices.