Search

Abstract Trajectory mapping techniques have widespread applications in diverse fields, including robotics, localization, smart environments, gaming, and tracking systems. However, existing free devices encounter challenges in representing trajectories, thereby limiting the effectiveness of applications such as robotics, localization, and tracking systems. The imprecise mappings generated by these methods lead to suboptimal performance and unreliable results. The proposed approach leverages WiFi sensing through channel state information (CSI), triangulation techniques, and a fine-tuning mechanism to enhance trajectory precision within indoor environment trajectory mapping. The proposed solution employs a domain adapter fine-tuning technique to enable location-independent tracking via CSI, minimizing errors. The use of CSI MIMO signals for trajectory mapping offers enhanced spatial resolution, robust multipath handling, and improved accuracy in tracking movement by leveraging multiple antenna channels and exploiting the rich information embedded in signal reflections and scattering, while triangulation aids in accurately determining the location of objects or targets. Furthermore, incorporating a fine-tuning mechanism refines the generated trajectories. The findings demonstrate substantial enhancements in mapping precision, with an accuracy of 95.5% in tracking 13 paths within the new domain. These results underscore the effectiveness of the proposed approach in overcoming the limitations of existing methods and achieving highly accurate trajectory mapping.

Abstract Fluorescence spectroscopy is a fundamental tool in life sciences and chemistry, with applications in environmental monitoring, food quality control, and biomedical diagnostics. However, analysis of spectroscopic data with deep learning, in particular of fluorescence excitation-emission matrices (EEMs), presents significant challenges due to the typically small and sparse datasets available. Furthermore, the analysis of EEMs is difficult due to their high dimensionality and overlapping spectral features. This study proposes a new approach that exploits domain adaptation with pretrained vision models, along with a novel interpretability algorithm to address these challenges. Thanks to specialised feature engineering of the neural networks described in this work, we are now able to provide deeper insights into the physico-chemical processes underlying the data. The proposed approach is demonstrated through the analysis of the oxidation process in extra virgin olive oil (EVOO), showing its effectiveness in predicting quality indicators and identifying the spectral bands and thus the molecules involved in the process. This work describes a significantly innovative approach to deep learning for spectroscopy, transforming it from a black box into a tool for understanding complex biological and chemical processes.

Trajectory mapping techniques have widespread applications in diverse fields, including robotics, localization, smart environments, gaming, and tracking systems. However, existing free devices encounter challenges in representing trajectories, thereby limiting the effectiveness of applications such as robotics, localization, and tracking systems. The imprecise mappings generated by these methods lead to suboptimal performance and unreliable results. The proposed approach leverages WiFi sensing through channel state information (CSI), triangulation techniques, and a fine-tuning mechanism to enhance trajectory precision within indoor environment trajectory mapping. The proposed solution employs a domain adapter fine-tuning technique to enable location-independent tracking via CSI, minimizing errors. The use of CSI MIMO signals for trajectory mapping offers enhanced spatial resolution, robust multipath handling, and improved accuracy in tracking movement by leveraging multiple antenna channels and exploiting the rich information embedded in signal reflections and scattering, while triangulation aids in accurately determining the location of objects or targets. Furthermore, incorporating a fine-tuning mechanism refines the generated trajectories. The findings demonstrate substantial enhancements in mapping precision, with an accuracy of 95.5% in tracking 13 paths within the new domain. These results underscore the effectiveness of the proposed approach in overcoming the limitations of existing methods and achieving highly accurate trajectory mapping. [ABSTRACT FROM AUTHOR]

Learning-based data compression methods have gained significant attention in recent years. Although these methods achieve higher compression ratios compared to traditional techniques, their slow processing times make them less suitable for compressing large datasets, and they are generally more effective for short texts rather than longer ones. In this study, MLMCompress, a word-based text compression method that can utilize any BERT masked language model is introduced. The performance of MLMCompress is evaluated using four BERT models: two large models and two smaller models referred to as "tiny". The large models are used without training, while the smaller models are fine-tuned. The results indicate that MLMCompress, when using the best-performing model, achieved 3838% higher compression ratios for English text and 42% higher compression ratios for multilingual text compared to NNCP, another learning-based method. Although the method does not yield better results than GPTZip, which has been developed in recent years, it achieves comparable outcomes while being up to 35 times faster in the worst-case scenario. Additionally, it demonstrated a 20% improvement in compression speed and a 180% improvement in decompression speed in the best case. Furthermore, MLMCompress outperforms traditional compression methods like Gzip and specialized short text compression methods such as Smaz and Shoco, particularly in compressing short texts, even when using smaller models. [ABSTRACT FROM AUTHOR]

Fluorescence spectroscopy is a fundamental tool in life sciences and chemistry, with applications in environmental monitoring, food quality control, and biomedical diagnostics. However, analysis of spectroscopic data with deep learning, in particular of fluorescence excitation-emission matrices (EEMs), presents significant challenges due to the typically small and sparse datasets available. Furthermore, the analysis of EEMs is difficult due to their high dimensionality and overlapping spectral features. This study proposes a new approach that exploits domain adaptation with pretrained vision models, along with a novel interpretability algorithm to address these challenges. Thanks to specialised feature engineering of the neural networks described in this work, we are now able to provide deeper insights into the physico-chemical processes underlying the data. The proposed approach is demonstrated through the analysis of the oxidation process in extra virgin olive oil (EVOO), showing its effectiveness in predicting quality indicators and identifying the spectral bands and thus the molecules involved in the process. This work describes a significantly innovative approach to deep learning for spectroscopy, transforming it from a black box into a tool for understanding complex biological and chemical processes. [ABSTRACT FROM AUTHOR]

Phishing and spam have been a cybersecurity threat with the majority of breaches resulting from these types of social engineering attacks. Therefore, detection has been a long‐standing challenge for both academic and industry researcher. New and innovative approaches are required to keep up with the growing sophistication of threat actors. One such illumination which has vast potential are large language models (LLM). LLM emerged and already demonstrated their potential to transform society and provide new and innovative approaches to solve well‐established challenges. Phishing and spam have caused financial hardships and lost time and resources to email users all over the world and frequently serve as an entry point for ransomware threat actors. While detection approaches exist, especially heuristic‐based approaches, LLMs offer the potential to venture into a new unexplored area for understanding and solving this challenge. LLMs have rapidly altered the landscape from business, consumers, and throughout academia and demonstrate transformational potential to profoundly impact the society. Based on this, applying these new and innovative approaches to email detection is a rational next step in academic research. In this work, we present IPSDM, an improved phishing spam detection model based on fine‐tuning the BERT family of models to specifically detect phishing and spam emails. We demonstrate our fine‐tuned version, IPSDM, is able to better classify emails in both unbalanced and balanced datasets. Moreover, IPSDM consistently outperforms the baseline models in terms of classification accuracy, precision, recall, and F1‐score, while concurrently mitigating overfitting concerns. [ABSTRACT FROM AUTHOR]

Gastrointestinal (GI) diseases are a major issue in the human digestive system. Therefore, many studies have explored the automatic classification of GI diseases to reduce the burden on clinicians and improve patient outcomes for both diagnosis and treatment purposes. Convolutional neural networks (CNNs) and Vision Transformers (ViTs) in deep learning approaches have become a popular research area for the automatic detection of diseases from medical images. This study evaluated the classification performance of thirteen different CNN models and two different ViT architectures on endoscopic images. The impact of transfer learning parameters on classification performance was also observed. The tests revealed that the classification accuracies of the ViT models were 91.25% and 90.50%, respectively. In contrast, the DenseNet201 architecture, with optimized transfer learning parameters, achieved an accuracy of 93.13%, recall of 93.17%, precision of 93.13%, and an F1 score of 93.11%, making it the most successful model among all the others. Considering the results, it is evident that a well-optimized CNN model achieved better classification performance than the ViT models. [ABSTRACT FROM AUTHOR]

The present research explores the pre-trained transformer's capability to provide conversational chatbot, using a niche approach of fine tuning the transformer on a custom dataset. Rags, Sentence Transformers, LLMs are all part of this suggested technique's implementation. This experiment examines the versatility of a pre-trained transformer and its ability to not only provide semantic context to highly sensitive conversation topics but also use its pre-training knowledge base to provide appropriate recommendations. The results are more promising to increase the nuances of a conversation chatbot. We give an original perspective on increasing the accuracy and effectiveness of an advanced chatbot that exceeds existing methodologies, revealing insight on the expanding landscape of artificial conversational chatbots using this complete methodology. [ABSTRACT FROM AUTHOR]

The key point of current research on few shot object detection based on meta learning is how to make better use of the information of support branch to help query branch to recognize novel objects more effectively. However, many current methods fuse the features from support branch and query branch in the depth direction, ignoring the spatial position relationship between features. Therefore, this study proposes non local support attention network. This method not only adds support information into the proposals features, but also fuses the support information with the features that fed into region proposal network. The spatial position relationship between features is considered at the same time. It also adds the information of negative supports to the detection module to help the model distinguish the objects from different categories. This method obtains good performance on base classes and novel classes of COCO2017 dataset, particularly under the case of incremental learning. Compared with method before improvement, 3.3/3.8/4.7 mAP is increased in AP/AP50/AP75 of novel classes. 2.7/0.5/3.3 mAP is increased in AP/AP50/AP75 of the base classes, and outperformed the performance of SOTA (Sort-Of-The Art) model DAnA (Dual-Awareness Attention) under the same setting. [ABSTRACT FROM AUTHOR]

This exploration acquaints a momentous methodology with custom chatbot improvement that focuses on proficiency close by viability. We accomplish this by joining three key innovations: LangChain, Retrieval Augmented Generation (RAG), and enormous language models (LLMs) tweaked with execution proficient strategies like LoRA and QLoRA. LangChain takes into consideration fastidious fitting of chatbots to explicit purposes, guaranteeing engaged and important collaborations with clients. RAG's web scratching capacities engage these chatbots to get to a tremendous store of data, empowering them to give exhaustive and enlightening reactions to requests. This recovered data is then decisively woven into reaction age utilizing LLMs that have been calibrated with an emphasis on execution productivity. This combination approach offers a triple advantage: further developed viability, upgraded client experience, and extended admittance to data. Chatbots become proficient at taking care of client questions precisely and productively, while instructive and logically pertinent reactions make a more regular and drawing in cooperation for clients. At last, web scratching enables chatbots to address a more extensive assortment of requests by conceding them admittance to a more extensive information base. By digging into the complexities of execution proficient LLM calibrating and underlining the basic job of web-scratched information, this examination offers a critical commitment to propelling custom chatbot plan and execution. The subsequent chatbots feature the monstrous capability of these advancements in making enlightening, easy to understand, and effective conversational specialists, eventually changing the manner in which clients cooperate with chatbots. [ABSTRACT FROM AUTHOR]

The NMOS cross-coupled LC Voltage-Controlled Oscillator (LCVCO) is considered an indispensable element in modern wireless communication systems. This research explores the design and implementation of a cross-coupled NMOS LCVCO using the UMC 65 nm CMOS process. The primary objective of this work is to achieve low Phase Noise for ka-band frequency applications with less power consumption. This article proposes Thermal and High frequency noise suppressed LCVCO (TH-LCVCO) technique. The TH-LCVCO technique incorporates a parallel capacitive filter to get rid of thermal noise at high frequencies and an LC tail filter to reduce the second-order harmonics. These techniques help to achieve low phase noise in all 16 tuning bands. The proposed TH-LCVCO technique consumes only 3.29 mW of power, achieving a frequency span from 32.17 GHz to 35.55 GHz with a Phase Noise of -112.6 dBc/Hz at 1 MHz and an Figure of Merit (FoM) of -197.300 dBc/Hz.

Background and Objectives: In recent years, various metaheuristic algorithms have gained popularity due to their effectiveness in solving complex optimization problems across diverse domains. These algorithms are now being utilized for an ever-increasing number of real-world applications. However, two important factors that significantly impact the performance of metaheuristic algorithms are understanding their behavior and fine-tuning their parameters. Deep understanding of an algorithm behavior assists in improving efficiency, while meticulous parameter calibration enhances optimization capability. Methods: In this study, a response surface methodology-based approach is proposed to analyze the behavior of optimization algorithms. This approach constructs a comprehensive model to determine parameter importance and interaction effects. Although applied to the Gravitational Search Algorithm, this methodology can serve as a generally applicable strategy to gain insights into any metaheuristic algorithm's functionality through quantitative and visual analysis. Results: Evaluation using 23 test functions exhibited that the technique identifies ideal parameter values and their comparative importance and interplays, enabling superior comprehension. Conclusion: The proposed framework utilizes informative modeling and multifaceted analysis to elucidate algorithm mechanics for more targeted calibration, thereby enhancing optimization performance. [ABSTRACT FROM AUTHOR]

The restricted access to data in healthcare facilities due to patient privacy and confidentiality policies has led to the application of general natural language processing (NLP) techniques advancing relatively slowly in the health domain. Additionally, because clinical data is unique to various institutions and laboratories, there are not enough standards and conventions for data annotation. In places without robust death registration systems, the cause of death (COD) is determined through a verbal autopsy (VA) report. A non-clinician field agent completes a VA report using a set of standardized questions as guide to identify the symptoms of a COD. The narrative text of the VA report is used as a case study to examine the difficulties of applying NLP techniques to the healthcare domain. This paper presents a framework that leverages knowledge across multiple domains via two domain adaptation techniques: feature extraction and fine-tuning. These techniques aim to improve VA text representations for COD classification tasks in the health domain. The framework is motivated by multi-step learning, where a final learning task is realized via a sequence of intermediate learning tasks. The framework builds upon the strengths of the Bidirectional Encoder Representations from Transformers (BERT) and Embeddings from Language Models (ELMo) models pretrained on the general English and biomedical domains. These models are employed to extract features from the VA narratives. Our results demonstrate improved performance when initializing the learning of BERT embeddings with ELMo embeddings. The benefit of incorporating character-level information for learning word embeddings in the English domain, coupled with word-level information for learning word embeddings in the biomedical domain, is also evident. [ABSTRACT FROM AUTHOR]

This study carries out a comprehensive comparison of fine-tuned GPT models (GPT-2, GPT-3, GPT-3.5) and LLaMA-2 models (LLaMA-2 7B, LLaMA-2 13B, LLaMA-2 70B) in text classification, addressing dataset sizes, model scales, and task diversity. Since its inception in 2018, the GPT series has been pivotal in advancing NLP, with each iteration introducing substantial enhancements. Despite its progress, detailed analyses, especially against competitive open-source models like the LLaMA-2 series in text classification, remain scarce. The current study fills this gap by fine-tuning these models across varied datasets, focusing on enhancing task-specific performance in hate speech and offensive language detection, fake news classification, and sentiment analysis. The learning efficacy and efficiency of the GPT and LLaMA-2 models were evaluated, providing a nuanced guide to choosing optimal models for NLP tasks based on architectural benefits and adaptation efficiency with limited data and resources. In particular, even with datasets as small as 1,000 rows per class, the F1 scores for the GPT-3.5 and LLaMA-2 models exceeded 0.9, reaching 0.99 with complete datasets. Additionally, the LLaMA-2 13B and 70B models outperformed GPT-3, demonstrating their superior efficiency and effectiveness in text classification. Both the GPT and LLaMA-2 series showed commendable performance on all three tasks, underscoring their ability to handle a diversity of tasks. Based on the size, performance, and resources required for fine-tuning the model, this study identifies LLaMA-2 13B as the most optimal model for NLP tasks. [ABSTRACT FROM AUTHOR]

This research applies the fine tuning attribute weighted naïve Bayes (FTAWNB) model using ordinal data. It is known that in previous research, the FTAWNB model outperformed its competitors on the dataset used. However, the FTAWNB model has not been applied in the mental health domain that uses ordinal data. Therefore, this research used the anxiety gamers dataset to test the fine-tuning attribute weighted naïve Bayes (FTAWNB) model. Anxiety disorders are mental health disorders that can indicate the emergence of a gaming disorder. Gamers can experience anxiety disorders classified into four classes, namely minimal, mild, moderate, and severe anxiety. Then compare the results by FTAWNB obtained with three other naïve Bayes algorithms, namely Gaussian naïve Bayes, multinomial naïve Bayes, and categorical naïve Bayes, using the same dataset. Model performance is measured based on accuracy, precision, recall, and processing time. The test results show that the FTAWNB outperforms the other three models' accuracy, precision, and recall, with an accuracy value of 99.22%. While the accuracy of Gaussian NB is 91.132%, Categorical is 91.592%, and multinomial naïve Bayes is 61.104%. However, the FTAWNB takes slightly longer than the other three models' processing time. The FTAWNB takes 0.07 seconds to build the model and 0.05 seconds to test the model on training data. [ABSTRACT FROM AUTHOR]

In the domain of general image forgery detection, a myriad of different classification solutions have been developed to distinguish a "tampered" image from a "pristine" image. In this work, we aim to develop a new method to tackle the problem of binary image forgery detection. Our approach builds upon the extensive training that state-of-the-art image classification models have undergone on regular images from the ImageNet dataset, and transfers that knowledge to the image forgery detection space. By leveraging transfer learning and fine tuning, we can fit state-of-the-art image classification models to the forgery detection task. We train the models on a diverse and evenly distributed image forgery dataset. With five models—EfficientNetB0, VGG16, Xception, ResNet50V2, and NASNet-Large—we transferred and adapted pre-trained knowledge from ImageNet to the forgery detection task. Each model was fitted, fine-tuned, and evaluated according to a set of performance metrics. Our evaluation demonstrated the efficacy of large-scale image classification models—paired with transfer learning and fine tuning—at detecting image forgeries. When pitted against a previously unseen dataset, the best-performing model of EfficientNetB0 could achieve an accuracy rate of nearly 89.7%. [ABSTRACT FROM AUTHOR]

Learning-based data compression methods have gained significant attention in recent years. Although these methods achieve higher compression ratios compared to traditional techniques, their slow processing times make them less suitable for compressing large datasets, and they are generally more effective for short texts rather than longer ones. In this study, MLMCompress, a word-based text compression method that can utilize any BERT masked language model is introduced. The performance of MLMCompress is evaluated using four BERT models: two large models and two smaller models referred to as “tiny”. The large models are used without training, while the smaller models are fine-tuned. The results indicate that MLMCompress, when using the best-performing model, achieved 3838% higher compression ratios for English text and 42% higher compression ratios for multilingual text compared to NNCP, another learning-based method. Although the method does not yield better results than GPTZip, which has been developed in recent years, it achieves comparable outcomes while being up to 35 times faster in the worst-case scenario. Additionally, it demonstrated a 20% improvement in compression speed and a 180% improvement in decompression speed in the best case. Furthermore, MLMCompress outperforms traditional compression methods like Gzip and specialized short text compression methods such as Smaz and Shoco, particularly in compressing short texts, even when using smaller models.

Assessments constitute a fundamental and inevitable component of any educational journey. Manual effort required for the evaluation of these assessments is very high. Automation of the evaluation process and grading helps in making the review process more efficient, objective, and scalable, thereby reducing the workload of human reviewers. Automating the grading process for multiple-choice and short-answer assessments is relatively straightforward, but it poses significant challenges when applied to the evaluation of formal languages, particularly in the context of mathematical assessments. In this paper a model that automatically evaluates and grades the Quadratic Equation problems is presented. The study is conducted using a manually curated dataset comprising 1200 solutions to various quadratic equation problems. Embeddings of the quadratic solutions are generated using Google's T5 Model. These embeddings are then used to train different traditional and ensembled machine learning models along with complex Deep learning models like LSTM and Bi LSTM. An in-depth analysis of the fine-tuned T5 model's performance, evaluating its effectiveness in comparison with the pretrained T5 model in automatic grading of quadratic equation problems has been explored. Fine-tuning significantly contributes to the reduction of error by 70% and a noticeable increase in the R2 value to 97%. [ABSTRACT FROM AUTHOR]

Paraphrase generation is a fundamental problem in natural language processing. Due to the significant success of transfer learning, the "pre-training → fine-tuning" approach has become the standard. However, popular general pre-training methods typically require extensive datasets and great computational resources, and the available pre-trained models are limited by fixed architecture and size. The authors have proposed a simple and efficient approach to pre-training specifically for paraphrase generation, which noticeably improves the quality of paraphrase generation and ensures substantial enhancement of general-purpose models. They have used existing public data and new data generated by large language models. The authors have investigated how this pre-training procedure impacts neural networks of various architectures and demonstrated its efficiency across all architectures. [ABSTRACT FROM AUTHOR]

Copyright of Informatica (03505596) is the property of Slovene Society Informatika and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

With the growth of the date fruit market, it becomes necessary to use artificial intelligence techniques to recognize dates categories. In this work, we use computer vision approaches to classify date fruits produced in Morocco according to their type. To do this, a dataset is constructed. This dataset contains images of seven types of date fruit that are the most popular in Morocco. Two computer vision approaches are compared and evaluated in terms of performance. Both approaches are based on transfer learning of a convolutional neural network CNN. These two approaches are standard feature extraction where deep features are used to train a machine learning classifier, and fine-tuning where we fit the pre-trained model to our dataset. The approaches used in this work achieve very good performance on our dataset, with a classification precision of 97 %. [ABSTRACT FROM AUTHOR]

Introduction: Insomnia causes serious adverse health effects and is estimated to affect 10-30% of the worldwide population. This study leverages personalized fine-tuned machine learning algorithms to detect insomnia risk based on questionnaire and longitudinal objective sleep data collected by a smart bed platform. Methods: Users of the Sleep Number smart bed were invited to participate in an IRB approved study which required them to respond to four questionnaires (which included the Insomnia Severity Index; ISI) administered 6 weeks apart from each other in the period from November 2021 to March 2022. For 1,489 participants who completed at least 3 questionnaires, objective data (which includes sleep/wake and cardio-respiratory metrics) collected by the platform were queried for analysis. An incremental, passive-aggressive machine learning model was used to detect insomnia risk which was defined by the ISI exceeding a given threshold. Three ISI thresholds (8, 10, and 15) were considered. The incremental model is advantageous because it allows personalized fine-tuning by adding individual training data to a generic model. Results: The generic model, without personalizing, resulted in an area under the receiving-operating curve (AUC) of about 0.5 for each ISI threshold. The personalized fine-tuning with the data of just five sleep sessions from the individual for whom the model is being personalized resulted in AUCs exceeding 0.8 for all ISI thresholds. Interestingly, no further AUC enhancements resulted by adding personalized data exceeding ten sessions. Discussion: These are encouraging results motivating further investigation into the application of personalized fine tuning machine learning to detect insomnia risk based on longitudinal sleep data and the extension of this paradigm to sleep medicine. [ABSTRACT FROM AUTHOR]

Citra generasi AI memiliki beragam manfaat yang signifikan, baik dalam bidang penelitian maupun industri. Namun, penggunaan citra generasi AI juga memiliki dampak negatif dalam konteks hukum, politik dan berbagai aspek lain dalam kehidupan. Penelitian ini menitik beratkan klasifikasi citra generasi AI yang dapat mendeteksi keaslian dari suatu citra. Metode yang diusulkan adalah menggunakan model residual network yang telah dilakukan fine tuning. Teknik fine tuning yang dilakukan meliputi penggunaan learning rate scheduler berbasis warm up yang diikuti dengan linear scheduler, akumulasi gradien, dan augmentasi citra. Penelitian menunjukkan bahwa model residual network 152 menghasilkan performa terbaik dengan f1 score 0.963 dan loss 0.08.

This paper introduces the “Couple Dialogue” dataset, specifically curated for conversational sentiment analysis in the Turkish language. It comprises 14,294 utterances from 118 dyadic conversations between couples, each annotated to capture sentiment transitions and interpersonal dynamics. Our study contrasts two distinct modeling frameworks—the Non-Contextual Model and the Contextual Model. The Non-Contextual Model analyzes utterances independently, typically overlooking the nuances of conversational dynamics and sentiment evolution. Within this framework, we conducted a detailed morphological analysis due to the agglutinative nature and rich morphological structure of the Turkish language, which included various word forms and negation morphemes crucial for sentiment representation. In contrast, the Contextual Model employs state-of-the-art Large Language Models (LLMs) such as BERT, GPT-3.5 Turbo, Llama 2, and GPT-4, alongside architectures like DialogueRNN. This model processes the sequential and relational aspects of dialogues through three approaches: prompt-based, fine-tuned, and embedding-based methods, particularly enhanced with fine-tuning and advanced embedding techniques (utilizing pre-trained and fine-tuned Turkish BERT). The Contextual Model substantially outperforms the Non-Contextual Model, showing a 9.98% improvement in Weighted F1 scores validated by statistical tests. Our work not only pioneers the use of LLMs in Turkish conversational sentiment analysis but also underscores the critical importance of contextual understanding in capturing complex emotional cues in couple interactions. This study sets a robust benchmark for future explorations into sentiment analysis within linguistically rich contexts.

Stereo camera self-calibration is a complex challenge in computer vision applications such as robotics, object tracking, surveillance and 3D reconstruction. To address this, we propose an efficient, fully automated End-To-End AI-Based system for automatic stereo camera self-calibration with varying intrinsic parameters, using only two images of any 3D scene. Our system combines deep convolutional neural networks (CNNs) with transfer learning techniques and fine-tuning. First, our end-to-end convolutional neural network optimized model begins by extracting matching points between a pair of stereo images. These matching points are then used, along with their 3D scene correspondences, to formulate a non-linear cost function. Direct optimization is subsequently performed to estimate the intrinsic camera parameters by minimizing this non-linear cost function. Following this initial optimization, a fine-tuning layer refines the intrinsic parameters for increased accuracy. Our hybrid approach is characterized by a special optimized architecture that leverages the strengths of end-to-end CNNs for image feature extraction and processing, as well as the pillars of our nonlinear cost function formulation and fine-tuning, to offer a robust and accurate method for stereo camera self-calibration. Extensive experiments on synthetic and real data demonstrate the superior performance of the proposed technique compared to traditional camera self-calibration methods in terms of precision and faster convergence.

This paper presents a comprehensive study on generating subjective inquiries for news media posts to empower public engagement with trending media topics. While previous studies primarily focused on factual and objective questions with explicit or implicit answers in the text, this research concentrates on automatically generating subjective questions to directly elicit personal preference from individuals based on a given text. The research methodology involves the application of fine-tuning techniques across multiple iterations of flan-T5 and GPT3 architectures for the task of Seq2Seq generation. This approach is meticulously evaluated using a custom dataset comprising 40,000 news articles along with human-generated questions. Furthermore, a comparative analysis is conducted using zero-shot prompting via GPT-3.5, juxtaposing the performance of fine-tuned models against a significantly larger language model. The study grapples with the inherent challenges tied to evaluating opinion-based question generation due to its subjective nature and the inherent uncertainty in determining answers. A thorough investigation and comparison of two transformer architectures are undertaken utilizing conventional lexical overlap metrics such as BLEU, ROUGE, and METEOR, alongside semantic similarity metrics encompassing BERTScore, BLEURT, and answerability scores such as QAScore, and RQUGE. The findings underscore the marked superiority of the flan-T5 model over GPT3, substantiated not only by quantitative metrics but also through human evaluations. The paper introduces Opinerium based on the open-source flan-T5-Large model, identified as the pacesetter in generating subjective questions. Additionally, we assessed all aforementioned metrics thoroughly by investigating the pairwise Spearman correlation analysis to identify robust metrics.

Drug Side Effects (DSE) are inconvenient and inadvertent retorts of the drugs. DSEs impact on public health and healthcare can prove costly. These DSEs can be an important factor in the failure/acceptance of drugs. Every approved drug should be either free from DSEs or these should be minor and reported properly. The drug discovery process should be capable of predicting and preventing these effects in advance. Previously, proposed studies for the prediction/prevention of DSEs utilized the features of 1D drug chemical structures or Natural Language Processing (NLP). Both these techniques required a complex transformation process. In this research authors have proposed a deep learning model, specifically using a transfer learning approach to predict DSEs directly from 2D chemical structure images, eliminating the need for the hefty transformation process of the NLP domain. For this study, a unique dataset is prepared that associates each image (taken from PubChem) with its specific side effects (SIDER). The results are evaluated using Accuracy, Precision, Recall and F-measure. The proposed model showed its dominance with an Accuracy of 73%, Precision of 83%, Recall of 73%, and an F1 score of 75%. The achieved results of the proposed model are compared against established transfer learning models like VGG16, DenseNet121 and some previously used traditional machine learning models like SVM and KNN. The collected results indicate a significant advancement in predicting drug side effects and offer a promising avenue for streamlining the drug development process.

Sleep analysis is widely and experimentally considered due to its importance to body health care. Since its sufficiency is essential for a healthy life, people often spend almost a third of their lives sleeping. In this case, a similar sleep pattern is not practiced by every individual, regarding pure healthiness or disorders such as insomnia, apnea, bruxism, epilepsy, and narcolepsy. Therefore, this study aims to determine the classification patterns of sleep stages, using big data for health care. This used a high-dimensional FFT extraction algorithm, as well as a feature importance and tuning classifier, to develop accurate classification. The results showed that the proposed method led to more accurate classification than previous techniques. This was because the previous experiments had been conducted with the feature selection model, with accuracy implemented as a performance evaluation. Meanwhile, the EEG Sleep Stages classification model in this present report was composed of the feature selection and importance of the extraction stage. The previous and present experiments also reached the highest values of accuracy, with the Random Forest and SVM models using 2000 and 3000 features (87.19% and 89.19%, respectively. In this article, we proposed an analysis that the feature importance subsequently influenced the model's accuracy. This was because the proposed method was easily fine-tuned and optimized for each subject to improve sensitivity and reduce false negative occurrences.

Recently, many open-source products have been used under the situations of general software development, because the cost saving and standardization. Therefore, many open-source products are gathering attention from many software development companies. Then, the reliability/quality of open-source products becomes very important factor for the software development. This paper focuses on the reliability/quality evaluation of open-source products. In particular, the large quantity fault data sets recorded on Bugzilla of open-source products is used in many open-source development projects. Then, the large amount of data sets of software faults is recorded on the Bugzilla. This paper proposes the reliability/quality evaluation approach based on the deep machine learning by using the large quantity fault data on the Bugzilla. Moreover, the large quantity fault data sets are analyzed by the deep machine learning based on the fine-tuning.

Leaf diseases of plants are common worldwide. Using image processing, farmers could spot diseases in pepper plants more rapidly and get advice from plant disease experts. In this paper, researchers developed a Transfer Learning classification model for bell pepper leaf disease, with the Transfer Learning model trained on images of healthy and diseased bell pepper leaves. Classification of healthy and diseased bell pepper leaves has been carried out, and fine-tuned Transfer Learning has been applied using several pre-trained CNN models. To achieve the best outcome, four pre-trained models, including MobileNet, VGG16, ResNetV250, and DenseNet121, and three Fully Connected (FC) layer architectures were tested. The Fully Connected (FC) layer with four Transfer Learning architectures achieved the best accuracy value of 99.33% on DenseNet121 architecture with one layer and Cohen’s Kappa value of 0.9865.

IntroductionInsomnia causes serious adverse health effects and is estimated to affect 10–30% of the worldwide population. This study leverages personalized fine-tuned machine learning algorithms to detect insomnia risk based on questionnaire and longitudinal objective sleep data collected by a smart bed platform.MethodsUsers of the Sleep Number smart bed were invited to participate in an IRB approved study which required them to respond to four questionnaires (which included the Insomnia Severity Index; ISI) administered 6 weeks apart from each other in the period from November 2021 to March 2022. For 1,489 participants who completed at least 3 questionnaires, objective data (which includes sleep/wake and cardio-respiratory metrics) collected by the platform were queried for analysis. An incremental, passive-aggressive machine learning model was used to detect insomnia risk which was defined by the ISI exceeding a given threshold. Three ISI thresholds (8, 10, and 15) were considered. The incremental model is advantageous because it allows personalized fine-tuning by adding individual training data to a generic model.ResultsThe generic model, without personalizing, resulted in an area under the receiving-operating curve (AUC) of about 0.5 for each ISI threshold. The personalized fine-tuning with the data of just five sleep sessions from the individual for whom the model is being personalized resulted in AUCs exceeding 0.8 for all ISI thresholds. Interestingly, no further AUC enhancements resulted by adding personalized data exceeding ten sessions.DiscussionThese are encouraging results motivating further investigation into the application of personalized fine tuning machine learning to detect insomnia risk based on longitudinal sleep data and the extension of this paradigm to sleep medicine.

In this study, the roles of shape and color features in metaphor generation for abstract images were investigated through simulations using retrained convolutional neural network (CNN) models based on the pretrained CNN model, AlexNet. A computational experiment was conducted using five types of retrained object recognition models: an object recognition model using the cleaned ILSVRC-2012 training dataset, one to recognize more shape features using edge-detected images, one to recognize fewer shape features using blurred images, one to recognize fewer color features using grayscale images, and one to recognize only shape features using Canny edge-detected images. The metaphors generated for abstract images were collected from behavioral data obtained in a psychological experiment aimed at investigating the neural mechanisms of metaphor generation for abstract images. In the computational experiment, the simulation results of the five models for abstract images were compared to examine how well they predicted the objects used in the metaphors generated for abstract images in the psychological experiment. The edge-only model using Canny edge-detected images and the color-inhibited model using grayscale images exhibited better performance in metaphor recognition for abstract images than the control condition. This indicates that shape features play a more important role than color features in metaphor generation for abstract images. Furthermore, because the Canny edge detection technique extracts only object outlines that can be regarded as the caricaturization of objects, the caricatured images, based on the shape features of the abstract images, likely influence object recognition for metaphor generation. [ABSTRACT FROM AUTHOR]

In the contemporary digital age, social media platforms like Facebook, Twitter, and YouTube serve as vital channels for individuals to express ideas and connect with others. Despite fostering increased connectivity, these platforms have inadvertently given rise to negative behaviors, particularly cyberbullying. While extensive research has been conducted on high-resource languages such as English, there is a notable scarcity of resources for low-resource languages like Bengali, Arabic, Tamil, etc., particularly in terms of language modeling. This study addresses this gap by developing a cyberbullying text identification system called BullyFilterNeT tailored for social media texts, considering Bengali as a test case. The intelligent BullyFilterNeT system devised overcomes Out-of-Vocabulary (OOV) challenges associated with non-contextual embeddings and addresses the limitations of context-aware feature representations. To facilitate a comprehensive understanding, three non-contextual embedding models GloVe, FastText, and Word2Vec are developed for feature extraction in Bengali. These embedding models are utilized in the classification models, employing three statistical models (SVM, SGD, Libsvm), and four deep learning models (CNN, VDCNN, LSTM, GRU). Additionally, the study employs six transformer-based language models: mBERT, bELECTRA, IndicBERT, XML-RoBERTa, DistilBERT, and BanglaBERT, respectively to overcome the limitations of earlier models. Remarkably, BanglaBERT-based BullyFilterNeT achieves the highest accuracy of 88.04% in our test set, underscoring its effectiveness in cyberbullying text identification in the Bengali language. [ABSTRACT FROM AUTHOR]

Deep-learning (DL) pickers have demonstrated superior performance in seismic phase picking compared to traditional pickers. DL pickers are extremely effective in processing large amounts of seismic data. Nevertheless, they encounter challenges when handling seismograms from different tectonic environments or source types, and even a slight change in the input waveform can considerably affect their consistency. Here, we fine-tuned a self-trained deep neural network picker using a small amount of local seismic data (26,875 three-component seismograms) recorded by regional seismic networks in South Korea. The self-trained model was developed using publicly available waveform datasets, comprising over two million three-component seismograms. The results revealed that the Korean-fine-tuned phase picker (KFpicker) effectively enhanced picking quality, even when applied to data that were not used during the fine-tuning process. When compared to the performance of the pre-trained model, this improvement was consistently observed regardless of variations in the positions of seismic phases in the input waveform, Furthermore, when the KFpicker predicted the phases for overlapping input windows and used the median value of probabilities as a threshold for phase detection, a considerable decrease was observed in the number of false picks. These findings indicate that fine-tuning a deep neural network using a small amount of local data can improve earthquake detection in the region of interest, while careful data augmentation can enhance the robustness of DL pickers against variations in the input window. The application of KFpicker to the 2016 Gyeongju earthquake sequence yielded approximately twice as many earthquakes compared to previous studies. Consequently, detailed and instantaneous statistical parameters of seismicity can be evaluated, making it possible to assess seismic hazard during an earthquake sequence. [ABSTRACT FROM AUTHOR]

The problem of classifying Woven Fabric Motifs through pattern recognition can be addressed using Convolutional Neural Networks (CNNs). Existing CNN architectures like VGG, ResNet, MobileNet, and DenseNet offer diverse propagation methods. These architectures, trained on datasets like imagenet, have demonstrated competence in solving large-scale classification tasks. The CNN model trained on the ImageNet dataset, hereinafter referred to as the pre-trained model, can be utilized to address the classification issue of NTT woven fabric motifs. This involves retraining the model using a new output layer and dataset, a method known as Transfer Learning. In addition to Transfer Learning, this research employs Fine Tuning, which entails retraining several classification layers. The pre-trained model used in this research is DenseNet121. This model was chosen because it does not require too much storage space and has good classification performance so that it can be embedded in smartphones. The results of this study indicate that of the three pre-trained models tested (DenseNet121, MobileNetV2, and ResNet50V2), the pre-trained Model DenseNet121 is the model that has the highest accuracy and the smallest loss, namely 92.58% accuracy and 29.62% Loss. Tests on mobile devices also show that from 130 test data, this model gets an accuracy of 99.23%. Overall, the classification model of NTT woven fabric motifs embedded in mobile devices can be used as an alternative to help the community or people who want to learn about NTT woven fabric motifs. [ABSTRACT FROM AUTHOR]