[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Sign in to use this feature.

Years

Between: -

Subjects

remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline

Journals

remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline

Article Types

Countries / Regions

remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline

Search Results (939)

Search Parameters:
Keywords = neural language model

Order results
Result details
Results per page
Select all
Export citation of selected articles as:
21 pages, 4816 KiB  
Article
Deep Learning-Based Postural Asymmetry Detection Through Pressure Mat
by Iker Azurmendi, Manuel Gonzalez, Gustavo García, Ekaitz Zulueta and Elena Martín
Appl. Sci. 2024, 14(24), 12050; https://doi.org/10.3390/app142412050 - 23 Dec 2024
Abstract
Deep learning, a subfield of artificial intelligence that uses neural networks with multiple layers, is rapidly changing healthcare. Its ability to analyze large datasets and extract relevant information makes it a powerful tool for improving diagnosis, treatment, and disease management. The integration of [...] Read more.
Deep learning, a subfield of artificial intelligence that uses neural networks with multiple layers, is rapidly changing healthcare. Its ability to analyze large datasets and extract relevant information makes it a powerful tool for improving diagnosis, treatment, and disease management. The integration of DL with pressure mats—which are devices that use pressure sensors to continuously and non-invasively monitor the interaction between patients and the contact surface—is a promising application. These pressure platforms generate data that can be very useful for detecting postural anomalies. In this paper we will discuss the application of deep learning algorithms in the analysis of pressure data for the detection of postural asymmetries in 139 patients aged 3 to 20 years. We investigated several main tasks: patient classification, hemibody segmentation, recognition of specific body parts, and generation of automated clinical reports. For this purpose, convolutional neural networks in their classification and regression modalities, the object detection algorithm YOLOv8, and the open language model LLaMa3 were used. Our results demonstrated high accuracy in all tasks: classification achieved 100% accuracy; hemibody division obtained an MAE of approximately 7; and object detection had an average accuracy of 70%. These results demonstrate the potential of this approach for monitoring postural and motor disabilities. By enabling personalized patient care, our methodology contributes to improved clinical outcomes and healthcare delivery. To our best knowledge, this is the first study that combines pressure images with multiple deep learning algorithms for the detection and assessment of postural disorders and motor disabilities in this group of patients. Full article
Show Figures

Figure 1

Figure 1
<p>Data preparation workflow [<a href="#B43-applsci-14-12050" class="html-bibr">43</a>].</p>
Full article ">Figure 2
<p>Dataset distribution: (<b>a</b>) by gender and (<b>b</b>) by postural asymmetry.</p>
Full article ">Figure 3
<p>Patient dataset features distribution: (<b>a</b>) height; (<b>b</b>) weight; and (<b>c</b>) age.</p>
Full article ">Figure 4
<p>Convolutional neural network architecture.</p>
Full article ">Figure 5
<p>Depiction of how YOLO works. Taken from the original YOLO article [<a href="#B52-applsci-14-12050" class="html-bibr">52</a>].</p>
Full article ">Figure 6
<p>Convolutional neural network input array representation.</p>
Full article ">Figure 7
<p>Distribution of postural asymmetries in training and validation sets.</p>
Full article ">Figure 8
<p>Model training and validation metrics: (<b>a</b>) accuracy and (<b>b</b>) loss.</p>
Full article ">Figure 9
<p>Body points for hemibody division.</p>
Full article ">Figure 10
<p>Training and validation metrics for the hemibody points prediction model: (<b>a</b>) six points prediction and (<b>b</b>) seven points prediction.</p>
Full article ">Figure 11
<p>Prediction results: (<b>a</b>) six points, prediction example 1; (<b>b</b>) six points, prediction example 2; (<b>c</b>) seven points, prediction example 1; and (<b>d</b>) seven points, prediction example 2.</p>
Full article ">Figure 12
<p>Hemibody pressure distribution examples charts: (<b>a</b>) patient that moves a lot during the test and (<b>b</b>) patient that barely moves during the test.</p>
Full article ">Figure 13
<p>Body parts recognition results: (<b>a</b>) example 1; (<b>b</b>) example 2; and (<b>c</b>) example 3.</p>
Full article ">Figure 14
<p>Body parts pressure percentage during tests: (<b>a</b>) patient that barely moves during the test and (<b>b</b>) patient that moves a lot during the test.</p>
Full article ">Figure 15
<p>Proposed report generation pipeline. Prompt definition obtained from [<a href="#B67-applsci-14-12050" class="html-bibr">67</a>].</p>
Full article ">
26 pages, 359 KiB  
Review
Opportunities and Challenges of Chatbots in Ophthalmology: A Narrative Review
by Mehmet Cem Sabaner, Rodrigo Anguita, Fares Antaki, Michael Balas, Lars Christian Boberg-Ans, Lorenzo Ferro Desideri, Jakob Grauslund, Michael Stormly Hansen, Oliver Niels Klefter, Ivan Potapenko, Marie Louise Roed Rasmussen and Yousif Subhi
J. Pers. Med. 2024, 14(12), 1165; https://doi.org/10.3390/jpm14121165 - 21 Dec 2024
Viewed by 344
Abstract
Artificial intelligence (AI) is becoming increasingly influential in ophthalmology, particularly through advancements in machine learning, deep learning, robotics, neural networks, and natural language processing (NLP). Among these, NLP-based chatbots are the most readily accessible and are driven by AI-based large language models (LLMs). [...] Read more.
Artificial intelligence (AI) is becoming increasingly influential in ophthalmology, particularly through advancements in machine learning, deep learning, robotics, neural networks, and natural language processing (NLP). Among these, NLP-based chatbots are the most readily accessible and are driven by AI-based large language models (LLMs). These chatbots have facilitated new research avenues and have gained traction in both clinical and surgical applications in ophthalmology. They are also increasingly being utilized in studies on ophthalmology-related exams, particularly those containing multiple-choice questions (MCQs). This narrative review evaluates both the opportunities and the challenges of integrating chatbots into ophthalmology research, with separate assessments of studies involving open- and close-ended questions. While chatbots have demonstrated sufficient accuracy in handling MCQ-based studies, supporting their use in education, additional exam security measures are necessary. The research on open-ended question responses suggests that AI-based LLM chatbots could be applied across nearly all areas of ophthalmology. They have shown promise for addressing patient inquiries, offering medical advice, patient education, supporting triage, facilitating diagnosis and differential diagnosis, and aiding in surgical planning. However, the ethical implications, confidentiality concerns, physician liability, and issues surrounding patient privacy remain pressing challenges. Although AI has demonstrated significant promise in clinical patient care, it is currently most effective as a supportive tool rather than as a replacement for human physicians. Full article
(This article belongs to the Section Methodology, Drug and Device Discovery)
16 pages, 2215 KiB  
Article
PoachNet: Predicting Poaching Using an Ontology-Based Knowledge Graph
by Naeima Hamed, Omer Rana, Pablo Orozco-terWengel, Benoît Goossens and Charith Perera
Sensors 2024, 24(24), 8142; https://doi.org/10.3390/s24248142 - 20 Dec 2024
Viewed by 248
Abstract
Poaching poses a significant threat to wildlife and their habitats, necessitating advanced tools for its prediction and prevention. Existing tools for poaching prediction face challenges such as inconsistent poaching data, spatiotemporal complexity, and translating predictions into actionable insights for conservation efforts. This paper [...] Read more.
Poaching poses a significant threat to wildlife and their habitats, necessitating advanced tools for its prediction and prevention. Existing tools for poaching prediction face challenges such as inconsistent poaching data, spatiotemporal complexity, and translating predictions into actionable insights for conservation efforts. This paper presents PoachNet, a novel predictive system that integrates deep learning with Semantic Web reasoning to infer poaching likelihood. Using elephant GPS data extracted from an ontology-based knowledge graph, PoachNet employs a sequential neural network to predict future movements, which are semantically modelled and incorporated into the graph. Semantic Web Rule Language (SWRL) is applied to infer poaching risk based on these geo-location predictions and poaching rule-based logic. By addressing spatiotemporal complexity and integrating predictions into an actionable semantic rule, PoachNet advances the field, with its geo-location prediction model outperforming state-of-the-art approaches. Full article
(This article belongs to the Section Sensor Networks)
Show Figures

Figure 1

Figure 1
<p>Multi-level Sankey Diagram to visualise Wildlife crime prediction authors, their approach, data used and the resultant product. Bakana et al. [<a href="#B33-sensors-24-08142" class="html-bibr">33</a>], Hofer et al. [<a href="#B32-sensors-24-08142" class="html-bibr">32</a>], Haas et al. [<a href="#B34-sensors-24-08142" class="html-bibr">34</a>], Edemacu et al. [<a href="#B42-sensors-24-08142" class="html-bibr">42</a>], Ferber et al. [<a href="#B43-sensors-24-08142" class="html-bibr">43</a>], Haas et al. [<a href="#B35-sensors-24-08142" class="html-bibr">35</a>], Gore et al. [<a href="#B37-sensors-24-08142" class="html-bibr">37</a>], Hamed et al. (PoachNet), Critchlow et al. [<a href="#B36-sensors-24-08142" class="html-bibr">36</a>], Gholami et al. [<a href="#B44-sensors-24-08142" class="html-bibr">44</a>], Kar et al. [<a href="#B40-sensors-24-08142" class="html-bibr">40</a>], Yang et al. [<a href="#B38-sensors-24-08142" class="html-bibr">38</a>], Fang et al. [<a href="#B45-sensors-24-08142" class="html-bibr">45</a>], Nguyen et al. [<a href="#B39-sensors-24-08142" class="html-bibr">39</a>], Gholami and McCarthy [<a href="#B41-sensors-24-08142" class="html-bibr">41</a>].</p>
Full article ">Figure 2
<p>The proposed ontology-based knowledge graph construction approach.</p>
Full article ">Figure 3
<p>The ontology-based knowledge graph lightweight version or conceptual model. FOO was visualized using the WebVOWL tool (version 1.1.7) available at (<a href="https://service.tib.eu/webvowl/" target="_blank">https://service.tib.eu/webvowl/</a>).</p>
Full article ">Figure 4
<p>PoachNet: End-to-end predictive framework featuring RDF data extraction and deep learning and showcasing the integration of the ontology-based knowledge graphs with deep learning. The framework consists of a sequential neural network for predicting an elephants future geo-location. The network comprises an input layer with a shape matching the dataset’s four features, followed by two hidden layers employing the Rectified Linear Unit (ReLU) activation function. The output layer uses a linear activation function. The model’s performance was assessed using the Root Mean Square Error (RMSE) metric, and accurate predictions were mapped back to their original RDF format.</p>
Full article ">Figure 5
<p>The map shows the geo-points (in green) intersection between the oil palm plantation (in blue) and the elephant movements (in red).</p>
Full article ">Figure 6
<p>The figure displays a 5 km buffer zone (in green) around each geographic feature of an oil palm plantation. These zones are overlaid with the original geographic features (shown with slight transparency) and the points from the elephant movements (in red).</p>
Full article ">Figure 7
<p>The chart visually compares the performance of PoachNet with other predictive models, specifically focusing on their Root Mean Square Error (RMSE) values, a standard measure of prediction accuracy. The other models include ‘Linear Regression’, ‘Polynomial Regression’, and ‘VAR’.</p>
Full article ">
19 pages, 5271 KiB  
Article
Design and Implementation of an Intelligent Web Service Agent Based on Seq2Seq and Website Crawler
by Mei-Hua Hsih, Jian-Xin Yang and Chen-Chiung Hsieh
Information 2024, 15(12), 818; https://doi.org/10.3390/info15120818 - 20 Dec 2024
Viewed by 238
Abstract
This paper proposes using a web crawler to organize website content as a dialogue tree in some domains. We build an intelligent customer service agent based on this dialogue tree for general usage. The encoder-decoder architecture Seq2Seq is used to understand natural language [...] Read more.
This paper proposes using a web crawler to organize website content as a dialogue tree in some domains. We build an intelligent customer service agent based on this dialogue tree for general usage. The encoder-decoder architecture Seq2Seq is used to understand natural language and then modified as a bi-directional LSTM to increase the accuracy of the polysemy cases. The attention mechanism is added in the decoder to improve the problem of accuracy decreasing as the sentence grows in length. We conducted four experiments. The first is an ablation experiment demonstrating that the Seq2Seq + Bi-directional LSTM + Attention mechanism is superior to LSTM, Seq2Seq, Seq2Seq + Attention mechanism in natural language processing. Using an open-source Chinese corpus for testing, the accuracy was 82.1%, 63.4%, 69.2%, and 76.1%, respectively. The second experiment uses knowledge of the target domain to ask questions. Five thousand data from Taiwan Water Supply Company were used as the target training data, and a thousand questions that differed from the training data but related to water were used for testing. The accuracy of RasaNLU and this study were 86.4% and 87.1%, respectively. The third experiment uses knowledge from non-target domains to ask questions and compares answers from RasaNLU with the proposed neural network model. Five thousand questions were extracted as the training data, including chat databases from eight public sources such as Weibo, Tieba, Douban, and other well-known social networking sites in mainland China and PTT in Taiwan. Then, 1000 questions from the same corpus that differed from the training data for testing were extracted. The accuracy of this study was 83.2%, which is far better than RasaNLU. It is confirmed that the proposed model is more accurate in the general field. The last experiment compares this study with voice assistants like Xiao Ai, Google Assistant, Siri, and Samsung Bixby. Although this study cannot answer vague questions accurately, it is more accurate in the trained application fields. Full article
Show Figures

Graphical abstract

Graphical abstract
Full article ">Figure 1
<p>Turing test. C asks A and B whether they are human [<a href="#B2-information-15-00818" class="html-bibr">2</a>].</p>
Full article ">Figure 2
<p>Proposed system architecture.</p>
Full article ">Figure 3
<p>Seq2Seq framework (<b>a</b>) without attention mechanism. (<b>b</b>) with an attention mechanism.</p>
Full article ">Figure 4
<p>Luong Attention.</p>
Full article ">Figure 5
<p>Bi-directional (<b>a</b>) RNN and (<b>b</b>) LSTM framework.</p>
Full article ">Figure 6
<p>They used model architecture for natural language processing.</p>
Full article ">Figure 7
<p>A part of the web content of the Taiwan Water Company.</p>
Full article ">Figure 8
<p>An example dialogue tree about the payment by credit card for the Taiwan Water Company.</p>
Full article ">Figure 9
<p>Chinese chat corpus.</p>
Full article ">Figure 10
<p>Annotation of the training data for RasaNLU.</p>
Full article ">Figure 11
<p>RasaNLU pipeline.</p>
Full article ">Figure 12
<p>A test example of the trained RasaNLU (read from left to right) for water fare query.</p>
Full article ">Figure 13
<p>(<b>a</b>) Training sample data. (<b>b</b>) LSTM training process.</p>
Full article ">Figure 14
<p>Some test results of Seq2Seq.</p>
Full article ">
18 pages, 8181 KiB  
Article
SMS Spam Detection System Based on Deep Learning Architectures for Turkish and English Messages
by Hakan Can Altunay and Zafer Albayrak
Appl. Sci. 2024, 14(24), 11804; https://doi.org/10.3390/app142411804 - 17 Dec 2024
Viewed by 365
Abstract
Short Message Service (SMS) still continues its existence despite the emergence of different messaging services. It plays a part in our lives as a communication service. Companies use SMS for advertisement purposes due to the fact that e-mail filtering systems have rooted, short [...] Read more.
Short Message Service (SMS) still continues its existence despite the emergence of different messaging services. It plays a part in our lives as a communication service. Companies use SMS for advertisement purposes due to the fact that e-mail filtering systems have rooted, short message systems are being undersold by the operators, and spam detection and blocking systems used for short messages are ineffective. Individuals falling victim to SMS spam messages sent by malevolent persons incur pecuniary and non-pecuniary losses. The aim of this study is to present a hybrid model proposal with the intention of detecting SMS spam messages. This detection model uses a gated recurrent unit (GRU) and convolutional neural network (CNN) as two deep learning methods. However, the fact that both algorithms require high memory capacities is a limitation. The design for this model was laid out by using two different datasets containing combined text messages written in the Turkish and English languages. The datasets used in the study are TurkishSMSCollection and the SMS Spam dataset from the UCI database. The testing process was performed on the dataset through benchmarking as well as other machine learning algorithms. It was revealed in the study that the hybrid CNN + GRU approach attained an accuracy of 99.07% by demonstrating a better performance compared to the other algorithms. Full article
(This article belongs to the Section Computing and Artificial Intelligence)
Show Figures

Figure 1

Figure 1
<p>Flow chart of the proposed model.</p>
Full article ">Figure 2
<p>Steps in word embedding.</p>
Full article ">Figure 3
<p>CNN + GRU model.</p>
Full article ">Figure 4
<p>GRU cell architecture.</p>
Full article ">Figure 5
<p>Results of algorithms.</p>
Full article ">Figure 6
<p>Confusion matrix values obtained during the testing stage.</p>
Full article ">Figure 7
<p>ROC of proposed model.</p>
Full article ">Figure 8
<p>Occurrence frequencies of words in the UCI SMS Spam dataset.</p>
Full article ">Figure 9
<p>Occurrence frequencies of words in the Turkish SMS Collection dataset.</p>
Full article ">
27 pages, 3027 KiB  
Article
DeepONet-Inspired Architecture for Efficient Financial Time Series Prediction
by Zeeshan Ahmad, Shudi Bao and Meng Chen
Mathematics 2024, 12(24), 3950; https://doi.org/10.3390/math12243950 - 16 Dec 2024
Viewed by 473
Abstract
Financial time series prediction is a fundamental problem in investment and risk management. Deep learning models, such as multilayer perceptrons, Convolutional Neural Networks (CNNs), and Long Short-Term Memory (LSTM), have been widely used in modeling time series data by incorporating historical information. Among [...] Read more.
Financial time series prediction is a fundamental problem in investment and risk management. Deep learning models, such as multilayer perceptrons, Convolutional Neural Networks (CNNs), and Long Short-Term Memory (LSTM), have been widely used in modeling time series data by incorporating historical information. Among them, LSTM has shown excellent performance in capturing long-term temporal dependencies in time-series data, owing to its enhanced internal memory mechanism. In spite of the success of these models, it is observed that in the presence of sharp changing points, these models fail to perform. To address this problem, we propose, in this article, an innovative financial time series prediction method inspired by the Deep Operator Network (DeepONet) architecture, which uses a combination of transformer architecture and a one-dimensional CNN network for processing feature-based information, followed by an LSTM based network for processing temporal information. It is therefore named the CNN–LSTM–Transformer (CLT) model. It not only incorporates external information to identify latent patterns within the financial data but also excels in capturing their temporal dynamics. The CLT model adapts to evolving market conditions by leveraging diverse deep-learning techniques. This dynamic adaptation of the CLT model plays a pivotal role in navigating abrupt changes in the financial markets. Furthermore, the CLT model improves the long-term prediction accuracy and stability compared with state-of-the-art existing deep learning models and also mitigates adverse effects of market volatility. The experimental results show the feasibility and superiority of the proposed CLT model in terms of prediction accuracy and robustness as compared to existing prediction models. Moreover, we posit that the innovation encapsulated in the proposed DeepONet-inspired CLT model also holds promise for applications beyond the confines of finance, such as remote sensing, data mining, natural language processing, and so on. Full article
(This article belongs to the Section Financial Mathematics)
Show Figures

Figure 1

Figure 1
<p>Analysis flowchart of the stock price prediction.</p>
Full article ">Figure 2
<p>Schematic diagram of the DeepONet.</p>
Full article ">Figure 3
<p>Proposed DeepONet-inspired architecture of the CLT model.</p>
Full article ">Figure 4
<p>Data sampling method.</p>
Full article ">Figure 5
<p>Loss curve on training and validation sets.</p>
Full article ">Figure 6
<p>Comparison of the prediction accuracy with varying number of nodes in a hidden layer.</p>
Full article ">Figure 7
<p>Comparison of the prediction accuracy with varying number of hidden layers.</p>
Full article ">Figure 8
<p>Comparison of the results produced by various models for the AEX.</p>
Full article ">Figure 9
<p>Comparison of the results produced by various models for the ATX.</p>
Full article ">Figure 10
<p>Model performance variation.</p>
Full article ">Figure A1
<p>Comparison of the results produced by various models for the following: (<b>a</b>) FCHI; (<b>b</b>) FTSE; (<b>c</b>) HSI; (<b>d</b>) JKSE; (<b>e</b>) KLSE; (<b>f</b>) OEX.</p>
Full article ">
15 pages, 2204 KiB  
Article
On the Functional Nature of Cognitive Systems
by Vincenzo Manca
Information 2024, 15(12), 807; https://doi.org/10.3390/info15120807 - 16 Dec 2024
Viewed by 445
Abstract
The functional nature of cognitive systems is outlined as a general conceptual model where typical notions of cognition are analyzed apart from the physical realization (biological or artificial) of such systems. The notion of function, one of the main logical bases of mathematics, [...] Read more.
The functional nature of cognitive systems is outlined as a general conceptual model where typical notions of cognition are analyzed apart from the physical realization (biological or artificial) of such systems. The notion of function, one of the main logical bases of mathematics, logic, linguistics, physics, and computer science, is shown to be a unifying concept in analyzing cognition components: learning, meaning, comprehension, language, knowledge, and consciousness are related to increasing levels in the functional organization of cognition. Full article
(This article belongs to the Section Information Applications)
Show Figures

Graphical abstract

Graphical abstract
Full article ">Figure 1
<p>A graph representation of the above FN expressed by a system of equations.</p>
Full article ">Figure 2
<p>FN holomorphy. Top: three weighted functions (bullets receive inputs and rectangles represent weights). Bottom: an FN is obtained by connecting the functions on the top, which provides a weighted function of the same kind as a single connected elements.</p>
Full article ">Figure 3
<p>An FN on top and its integration with learning FN on bottom. Integration is represented at two levels, employing a reverse network of nodes that are arrow bridges of original FN.</p>
Full article ">Figure 4
<p>An FN and a meta-function K adjusting a weight according to an input error (between the computed function and a target function).</p>
Full article ">Figure 5
<p>The translation of the meta-function K into a function providing the same effect according to a bridge mechanism.</p>
Full article ">Figure 6
<p>The inputs I1, I2, and I3 of F-G-H are sent to the FN with six Id functions (Id is the identity), where three meta-functions update three weights. When input 1 is given to the synapses on the bottom, G sends to F the same value generated by I1, I2, and I3. In other words, the FN on the bottom memorizes inputs of F-G-H as weights (those between the pairs of Id functions). This representation individuates a memory mechanism transforming input values into weights, where meta-functions are essential (weights indicated by slim rectangles have value 1).</p>
Full article ">
16 pages, 240 KiB  
Article
A Comparative Study of Sentiment Analysis on Customer Reviews Using Machine Learning and Deep Learning
by Logan Ashbaugh and Yan Zhang
Computers 2024, 13(12), 340; https://doi.org/10.3390/computers13120340 - 15 Dec 2024
Viewed by 498
Abstract
Sentiment analysis is a key technique in natural language processing that enables computers to understand human emotions expressed in text. It is widely used in applications such as customer feedback analysis, social media monitoring, and product reviews. However, sentiment analysis of customer reviews [...] Read more.
Sentiment analysis is a key technique in natural language processing that enables computers to understand human emotions expressed in text. It is widely used in applications such as customer feedback analysis, social media monitoring, and product reviews. However, sentiment analysis of customer reviews presents unique challenges, including the need for large datasets and the difficulty in accurately capturing subtle emotional nuances in text. In this paper, we present a comparative study of sentiment analysis on customer reviews using both deep learning and traditional machine learning techniques. The deep learning models include Convolutional Neural Network (CNN) and Recursive Neural Network (RNN), while the machine learning methods consist of Logistic Regression, Random Forest, and Naive Bayes. Our dataset is composed of Amazon product reviews, where we utilize the star rating as a proxy for the sentiment expressed in each review. Through comprehensive experiments, we assess the performance of each model in terms of accuracy and effectiveness in detecting sentiment. This study provides valuable insights into the strengths and limitations of both deep learning and traditional machine learning approaches for sentiment analysis. Full article
16 pages, 538 KiB  
Article
What ChatGPT Has to Say About Its Topological Structure: The Anyon Hypothesis
by Michel Planat and Marcelo Amaral
Mach. Learn. Knowl. Extr. 2024, 6(4), 2876-2891; https://doi.org/10.3390/make6040137 - 15 Dec 2024
Viewed by 494
Abstract
Large language models (LLMs) achieve remarkable predictive capabilities but remain opaque in their internal reasoning, creating a pressing need for more interpretable artificial intelligence. Here, we propose bridging this explanatory gap by drawing on concepts from topological quantum computing (TQC), specifically the anyonic [...] Read more.
Large language models (LLMs) achieve remarkable predictive capabilities but remain opaque in their internal reasoning, creating a pressing need for more interpretable artificial intelligence. Here, we propose bridging this explanatory gap by drawing on concepts from topological quantum computing (TQC), specifically the anyonic frameworks arising from SU(2)k theories. Anyons interpolate between fermions and bosons, offering a mathematical language that may illuminate the latent structure and decision-making processes within LLMs. By examining how these topological constructs relate to token interactions and contextual dependencies in neural architectures, we aim to provide a fresh perspective on how meaning and coherence emerge. After eliciting insights from ChatGPT and exploring low-level cases of SU(2)k models, we argue that the machinery of modular tensor categories and topological phases could inform more transparent, stable, and robust AI systems. This interdisciplinary approach suggests that quantum-theoretic principles may underpin a novel understanding of explainable AI. Full article
16 pages, 1308 KiB  
Article
Evaluating DL Model Scaling Trade-Offs During Inference via an Empirical Benchmark Analysis
by Demetris Trihinas, Panagiotis Michael and Moysis Symeonides
Future Internet 2024, 16(12), 468; https://doi.org/10.3390/fi16120468 - 13 Dec 2024
Viewed by 564
Abstract
With generative Artificial Intelligence (AI) capturing public attention, the appetite of the technology sector for larger and more complex Deep Learning (DL) models is continuously growing. Traditionally, the focus in DL model development has been on scaling the neural network’s foundational structure to [...] Read more.
With generative Artificial Intelligence (AI) capturing public attention, the appetite of the technology sector for larger and more complex Deep Learning (DL) models is continuously growing. Traditionally, the focus in DL model development has been on scaling the neural network’s foundational structure to increase computational complexity and enhance the representational expressiveness of the model. However, with recent advancements in edge computing and 5G networks, DL models are now aggressively being deployed and utilized across the cloud–edge–IoT continuum for the realization of in situ intelligent IoT services. This paradigm shift introduces a growing need for AI practitioners, as a focus on inference costs, including latency, computational overhead, and energy efficiency, is long overdue. This work presents a benchmarking framework designed to assess DL model scaling across three key performance axes during model inference: classification accuracy, computational overhead, and latency. The framework’s utility is demonstrated through an empirical study involving various model structures and variants, as well as publicly available datasets for three popular DL use cases covering natural language understanding, object detection, and regression analysis. Full article
Show Figures

Figure 1

Figure 1
<p>High-level overview of a deep neural network.</p>
Full article ">Figure 2
<p>Pipeline of performance evaluation trade-offs.</p>
Full article ">Figure 3
<p>Inference quality (classification accuracy and MSE) with respect to model complexity. The presented plots include: (<b>a</b>) BERT model variants, (<b>b</b>) EfficientNet model variants, and (<b>c</b>) MLP-Regression model variants.</p>
Full article ">Figure 4
<p>Computational overhead of inference with respect to model complexity. The presented plots include: (<b>a</b>) BERT model variants, (<b>b</b>) EfficientNet model variants, and (<b>c</b>) MLP-Regression model variants.</p>
Full article ">Figure 5
<p>Inference latency with respect to model complexity. The presented plots include: (<b>a</b>) BERT model variants, (<b>b</b>) EfficientNet model variants, and (<b>c</b>) MLP-Regression model variants.</p>
Full article ">
22 pages, 838 KiB  
Article
MediScan: A Framework of U-Health and Prognostic AI Assessment on Medical Imaging
by Sibtain Syed, Rehan Ahmed, Arshad Iqbal, Naveed Ahmad and Mohammed Ali Alshara
J. Imaging 2024, 10(12), 322; https://doi.org/10.3390/jimaging10120322 - 13 Dec 2024
Viewed by 789
Abstract
With technological advancements, remarkable progress has been made with the convergence of health sciences and Artificial Intelligence (AI). Modern health systems are proposed to ease patient diagnostics. However, the challenge is to provide AI-based precautions to patients and doctors for more accurate risk [...] Read more.
With technological advancements, remarkable progress has been made with the convergence of health sciences and Artificial Intelligence (AI). Modern health systems are proposed to ease patient diagnostics. However, the challenge is to provide AI-based precautions to patients and doctors for more accurate risk assessment. The proposed healthcare system aims to integrate patients, doctors, laboratories, pharmacies, and administrative personnel use cases and their primary functions onto a single platform. The proposed framework can also process microscopic images, CT scans, X-rays, and MRI to classify malignancy and give doctors a set of AI precautions for patient risk assessment. The proposed framework incorporates various DCNN models for identifying different forms of tumors and fractures in the human body i.e., brain, bones, lungs, kidneys, and skin, and generating precautions with the help of the Fined-Tuned Large Language Model (LLM) i.e., Generative Pretrained Transformer 4 (GPT-4). With enough training data, DCNN can learn highly representative, data-driven, hierarchical image features. The GPT-4 model is selected for generating precautions due to its explanation, reasoning, memory, and accuracy on prior medical assessments and research studies. Classification models are evaluated by classification report (i.e., Recall, Precision, F1 Score, Support, Accuracy, and Macro and Weighted Average) and confusion matrix and have shown robust performance compared to the conventional schemes. Full article
Show Figures

Figure 1

Figure 1
<p>Graphical scheme of the system architecture.</p>
Full article ">Figure 2
<p>Graphical scheme of use cases in the proposed framework.</p>
Full article ">Figure 3
<p>Graphical illustration of proposed AI bones fracture detection model.</p>
Full article ">Figure 4
<p>Graphical illustration of the proposed AI lung cancer detection model.</p>
Full article ">Figure 5
<p>Graphical illustration of the proposed AI brain tumor detection model.</p>
Full article ">Figure 6
<p>Graphical illustration of the proposed AI skin cancer detection model.</p>
Full article ">Figure 7
<p>Graphical illustration of the proposed AI kidney malignancy detection model.</p>
Full article ">Figure 8
<p>Graphical illustration of the proposed GPT-4 model system integration.</p>
Full article ">Figure 9
<p>Graphical illustration of the confusion matrix for Bone Fracture recognition; Lung Tumor recognition; Brain Tumor detection; Skin Lesion identification; and Renal Malignancy recognition AI model.</p>
Full article ">Figure 10
<p>Graphical illustration of accuracy graph for Bone Fracture recognition; Lung Tumor recognition; Brain Tumor detection; Skin Lesion identification; and Renal Malignancy recognition AI model.</p>
Full article ">
17 pages, 1870 KiB  
Article
Semantically-Enhanced Feature Extraction with CLIP and Transformer Networks for Driver Fatigue Detection
by Zhen Gao, Xiaowen Chen, Jingning Xu, Rongjie Yu, Heng Zhang and Jinqiu Yang
Sensors 2024, 24(24), 7948; https://doi.org/10.3390/s24247948 - 12 Dec 2024
Viewed by 410
Abstract
Drowsy driving is a leading cause of commercial vehicle traffic crashes. The trend is to train fatigue detection models using deep neural networks on driver video data, but challenges remain in coarse and incomplete high-level feature extraction and network architecture optimization. This paper [...] Read more.
Drowsy driving is a leading cause of commercial vehicle traffic crashes. The trend is to train fatigue detection models using deep neural networks on driver video data, but challenges remain in coarse and incomplete high-level feature extraction and network architecture optimization. This paper pioneers the use of the CLIP (Contrastive Language-Image Pre-training) model for fatigue detection. And by harnessing the power of a Transformer architecture, sophisticated and long-term temporal features are adeptly extracted from video sequences, paving the way for more nuanced and accurate fatigue analysis. The proposed CT-Net (CLIP-Transformer Network) achieves an AUC (Area Under the Curve) of 0.892, a 36% accuracy improvement over the prevalent CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory) end-to-end model, reaching state-of-the-art performance. Experiments show that the CLIP pre-trained model more accurately extracts facial and behavioral features from driver video frames, improving the model’s AUC by 7% over the ImageNet-based pre-trained model. Moreover, compared with LSTM, the Transformer more flexibly captures long-term dependencies among temporal features, further enhancing the model’s AUC by 4%. Full article
(This article belongs to the Section Intelligent Sensors)
Show Figures

Figure 1

Figure 1
<p>CT-Net architecture.</p>
Full article ">Figure 2
<p>Loss function curve of the CT-Net model and the baseline model.</p>
Full article ">Figure 3
<p>Semantic interpretation utilizing eye state-related prompt. (<b>a</b>) The change in probability of an eye-closing event that occurs during normal driving. (<b>b</b>) The change in probability of an eye-closing event that occurs during fatigued driving. (<b>c</b>) The change in probability of an eye-closing event that occurs during highly fatigued driving.</p>
Full article ">Figure 4
<p>Semantic interpretation utilizing mouth state-related prompt. (<b>a</b>) The change in probability of a yawn event that occurs during normal driving. (<b>b</b>) The change in probability of a yawn event that occurs during fatigued driving. (<b>c</b>) The change in probability of a yawn event that occurs during highly fatigued driving.</p>
Full article ">Figure 5
<p>Semantic interpretation utilizing some behavior-related prompt.</p>
Full article ">Figure 6
<p>Training and validation loss curve of three experiments.</p>
Full article ">
13 pages, 3483 KiB  
Article
Classification of English Words into Grammatical Notations Using Deep Learning Technique
by Muhammad Imran, Sajjad Hussain Qureshi, Abrar Hussain Qureshi and Norah Almusharraf
Information 2024, 15(12), 801; https://doi.org/10.3390/info15120801 - 11 Dec 2024
Viewed by 408
Abstract
The impact of artificial intelligence (AI) on English language learning has become the center of attention in the past few decades. This study, with its potential to transform English language instruction and offer various instructional approaches, provides valuable insights and knowledge. To fully [...] Read more.
The impact of artificial intelligence (AI) on English language learning has become the center of attention in the past few decades. This study, with its potential to transform English language instruction and offer various instructional approaches, provides valuable insights and knowledge. To fully grasp the potential advantages of AI, more research is needed to improve, validate, and test AI algorithms and architectures. Grammatical notations provide a word’s information to the readers. If a word’s images are properly extracted and categorized using a CNN, it can help non-native English speakers improve their learning habits. The classification of parts of speech into different grammatical notations is the major problem that non-native English learners face. This situation stresses the need to develop a computer-based system using a machine learning algorithm to classify words into proper grammatical notations. A convolutional neural network (CNN) was applied to classify English words into nine classes: noun, pronoun, adjective, determiner, verb, adverb, preposition, conjunction, and interjection. A simulation of the selected model was performed in MATLAB. The model achieved an overall accuracy of 97.22%. The CNN showed 100% accuracy for pronouns, determiners, verbs, adverbs, and prepositions; 95% for nouns, adjectives, and conjunctions; and 90% for interjections. The significant results (p < 0.0001) of the chi-square test supported the use of the CNN by non-native English learners. The proposed approach is an important source of word classification for non-native English learners by putting the word image into the model. This not only helps beginners in English learning but also helps in setting standards for evaluating documents. Full article
(This article belongs to the Special Issue Applications of Machine Learning and Convolutional Neural Networks)
Show Figures

Figure 1

Figure 1
<p>Conceptual design of the proposed model.</p>
Full article ">Figure 2
<p>The proposed architecture of the lightweight model.</p>
Full article ">Figure 3
<p>Feature mapping and max pooling technique of CNN model.</p>
Full article ">Figure 4
<p>(<b>a</b>) Training and validation accuracy of the proposed model. (<b>b</b>) Training and validation loss of the proposed model.</p>
Full article ">Figure 5
<p>Confusion Matrix of parts of speech into nine classes.</p>
Full article ">Figure 6
<p>Results of model predictions.</p>
Full article ">Figure 7
<p>Comparison of grammatical notation recognition.</p>
Full article ">
20 pages, 19406 KiB  
Article
Research on the Application of Topic Models Based on Geological Disaster Information Mining
by Gang Cheng, Qinliang You, Gangqiang Li, Youcai Li, Daisong Yang, Jinghong Wu and Yaxi Wu
Information 2024, 15(12), 795; https://doi.org/10.3390/info15120795 - 10 Dec 2024
Viewed by 461
Abstract
Geological disasters, as a common occurrence, have a serious impact on social development in terms of their frequency of occurrence, disaster effects, and resulting losses. To effectively reduce the casualties, property losses, and social effects caused by various disasters, it is necessary to [...] Read more.
Geological disasters, as a common occurrence, have a serious impact on social development in terms of their frequency of occurrence, disaster effects, and resulting losses. To effectively reduce the casualties, property losses, and social effects caused by various disasters, it is necessary to conduct real-time monitoring and early warning of various geological disaster risks. With the growing development of the information age, public attention to disaster relief, casualties, social impact effects, and other related situations has been increasing. Since social media platforms such as Weibo and Twitter contain a vast amount of real-time data related to disaster information before and after a disaster occurs, scientifically and effectively utilizing these data can provide sufficient and reliable information support for disaster relief, post-disaster recovery, and public appeasement efforts. As one of the techniques in natural language processing, the topic model can achieve precise mining and intelligent analysis of valuable information from massive amounts of data on social media to achieve rapid use of thematic models for disaster analysis after a disaster occurs, providing reference for post-disaster-rescue-related work. Therefore, this article first provides an overview of the development process of the topic model. Secondly, based on the technology utilized, the topic models were roughly classified into three categories: traditional topic models, word embedding-based topic models, and neural network-based topic models. Finally, taking the disaster data of “Dongting Lake breach” in Hunan, China as the research object, the application process and effectiveness of the topic model in urban geological disaster information mining were systematically introduced. The research results provide important references for the further practical innovation and expansion of the topic model in the field of disaster information mining. Full article
(This article belongs to the Section Information Processes)
Show Figures

Figure 1

Figure 1
<p>Sites of various geological disasters. (<b>a</b>) Collapse of waste soil heap in Myanmar’s Vika jade mining area; (<b>b</b>) Landslides in certain regions of India; (<b>c</b>) India and Pakistan hit by mudslides and landslides; (<b>d</b>) Mountain flood and debris flow in Heziping Village, Xi’an, China.</p>
Full article ">Figure 2
<p>Distribution of various geological disasters in China in 2022.</p>
Full article ">Figure 3
<p>Classification of topic models.</p>
Full article ">Figure 4
<p>The simulation login process for Sina Weibo.</p>
Full article ">Figure 5
<p>Data cleaning process.</p>
Full article ">Figure 6
<p>Chinese word segmentation algorithm and model.</p>
Full article ">Figure 7
<p>Aerial view of the dike breach in Dongting Lake.</p>
Full article ">Figure 8
<p>“Dongting Lake dike breach” corpus.</p>
Full article ">Figure 9
<p>Perplexity curve of “Dongting Lake dike breach” topics.</p>
Full article ">Figure 10
<p>Visualization of topic–word distribution results [<a href="#B54-information-15-00795" class="html-bibr">54</a>,<a href="#B55-information-15-00795" class="html-bibr">55</a>].</p>
Full article ">Figure 11
<p>The partial word cloud for “Dongting Lake dike breach” topics.</p>
Full article ">Figure 12
<p>Popularity distribution of Weibo disaster data.</p>
Full article ">
19 pages, 450 KiB  
Article
Detecting Aggression in Language: From Diverse Data to Robust Classifiers
by Aleksander Wawer, Agnieszka Mykowiecka and Bartosz Żuk
Electronics 2024, 13(24), 4857; https://doi.org/10.3390/electronics13244857 - 10 Dec 2024
Viewed by 460
Abstract
The automatic detection of aggressive language is a difficult challenge. Currently, three datasets are available in Polish, enabling the training of machine learning models to recognise different types of linguistic aggression. In this paper, we address the issues of the transferability of knowledge [...] Read more.
The automatic detection of aggressive language is a difficult challenge. Currently, three datasets are available in Polish, enabling the training of machine learning models to recognise different types of linguistic aggression. In this paper, we address the issues of the transferability of knowledge between datasets and training a single model that works best on all types of aggression. Due to data imbalance, we experiment with two loss functions dedicated to training on imbalanced data: Weighted Cross-Entropy and Focal loss. Using the Polish language HerBERT model, we present the results of experiments in the Cross-dataset scenario and the model results using the combined data. Our results show that (1) combining diverse types of linguistic aggression during training leads to a better-performing classifier and (2) Weighted Cross-Entropy outperforms other tested loss functions. Full article
(This article belongs to the Special Issue Emerging Theory and Applications in Natural Language Processing)
Show Figures

Figure 1

Figure 1
<p>Length and label distributions: (<b>a</b>) box plot of utterances’ length distribution, measured in tokens and grouped by class label, in each dataset (for readability we omit outliers), (<b>b</b>) bar plot of class label proportions in each dataset.</p>
Full article ">Figure 2
<p>Quality and complexity metrics: (<b>a</b>) bar plot of percentage of OOV tokens in each dataset, (<b>b</b>) bar plot showing syntactic complexity metrics in each dataset.</p>
Full article ">Figure 3
<p>Bar plot comparing F1 scores for selected baseline models from <a href="#electronics-13-04857-t004" class="html-table">Table 4</a> across three datasets.</p>
Full article ">Figure 4
<p>Confusion matrices illustrating the performance of six models. The first four matrices represent models trained on individual datasets: Cyberbullying, Hate Speech, Aggression-1:1, and Aggression-1:10. The last two matrices show the results for the binary and non-binary models trained on the Mixed-dataset. The numbers in each cell indicate, from top to bottom, the number of texts in the set and the number of texts normalised by the total size of the dataset, and, at the bottom (in parentheses), the number of texts normalised by the number of texts belonging to the given class within the set (in square brackets).</p>
Full article ">
Back to TopTop