8333 research outputs found
Sort by
Enhancing Virtual Machine Migration through the utilization of Reinforcement Learning and Apache Kafka Messaging Services
No full textThis research introduces an innovative approach to perform VM migration with the use of Reinforcement Learning (RL) and Apache Kafka Messaging Services. The system employs an RL agent for migration decision making, and offers opportunities to monitor actual usage data like the CPU, memory and network usage. This metrics can be streamed to the Apache Kafka in low latency high through put ensuring rich data transmission. Rewards are given to the RL agent for migration actions depending on the efficiency of the performed actions. Experiment findings prove that the utilization of RL-based systems contributes to the enhancement of the general effectiveness of decision- making procedures, the increase of the efficiency of the systems as a whole as well as the effective utilization of resources in comparison with other approaches. The architecture of Kafka is distributed and this characteristic enables it to work at scalable modes and manage increasing data volumes without any loss of performance. The adaptability of RL agent allows continuous learning and optimization of migration strategies thus providing a dynamic and efficient solution for modern data centers. This research demonstrates the potential of combining machine learning with real time data streaming to improve VM migration process offering more responsive cloud infrastructure management
Secure And Verifiable Cloud Based Data Sharing For Law Enforcement With Sensitive Information And Sharing
No full textIn the field of the law enforcement the managing and sharing of sensitive data specially criminal records containing Personally Identifiable Information (PII) which is very sensitive and is of most importance. This Secure and Verifiable Cloud-Based Data Sharing for Law Enforcement with Sensitive Information and Encryption project addresses the critical need for enhanced data security , done by developing a robust system for detecting, encrypting, and securely sharing PII within the criminal datasets. Using AWS cloud services including S3 for data storage, Lambda for event driven orchestration and SageMaker for automated processing, this project ensures that the sensitive information is protected throughout the entire lifecycle of data.
The law enforcement data encryption project employs the Fernet symmetric encryption algorithm in order to secure the PII, ensuring that data is only accessible to authorized individuals and not everyone. The system is designed in a way that it automatically triggers data processing workflows when new datasets is uploaded to the S3,thereby allowing for real-time encryption and secure storage. Docker containers are utilized within the SageMaker environment to provide the consistent runtime ensuring that the processing tasks are performed efficiently and securely.
This project majorly focuses on providing a broad and scalable solution for the secure management of criminal datasets in the cloud environments, thereby reducing the risk of breaching of data and unauthorized access to it. By facilitating secure data sharing among the law enforcement agencies and the legal organizations the system not only protects sensitive information but it also supports effective collaboration across multiple stakeholders involved
Leadership for inclusion: Navigating the evolving landscape of the FET sector in Ireland
No full textThe Further Education and Training (FET) sector in Ireland offers a significant level of diversity, in student population, level of study, and form of delivery. As such, inclusive provision is crucial to the sector’s success as a viable learning pathway. Key to this process is the work of leaders, as there is ample evidence of their impact on effective inclusive policy and practice. The vast majority of existing research in this area has focused on primary, post-primary and higher education sectors with an evident lack of such work in FET. This project addresses this gap, illuminating perspectives and practices around leadership for inclusion in Irish FET settings, based on first-hand accounts from senior leaders. Five leaders in a range of FET settings participated in an exploratory qualitative inquiry with two researchers. The findings reveal a common conceptualisation of inclusion as rights-based and far-reaching where leaders are evidently committed to fully including all members of their respective populations. Leaders acknowledge their own role in modelling inclusive practice, but somewhat dichotomously highlight a lack of visibility around inclusive teaching and learning. Finally, they acknowledge that FET’s diminished status in comparison to other sectors has resulted in difficulties around gaining and employing supports for learners, but they also demonstrate a belief that this same status has undergone a sense of renewal in recent times
Data Drift for Automatic FAIR-compliant Dataset Versioning in Large Repositories
No full textConstrued as a shift in the distribution or structure of data over time, data drift can adversely affect the performance of machine learning models and data-driven decisions. This study examines two data drift metrics, denoted as d E,PCA and d E,AE , that are derived from unsupervised ML models: the reconstruction error-based metrics of Principal Component Analysis (PCA) and Autoencoders (AE). To investigate the robustness of these metrics, we have systematically accessed time-series datasets from the European Data Portal. Our experiments have examined data versioning through three basic events: creation, update, and deletion. The results are summarised and aggregated for all datasets, and unsupervised analysis based on Robust PCA and AE has been performed to examine patterns within the impact of dataset characteristics on data drift detection and computational efficiency. Our results indicate that both metrics aligned closely in performance with new records, suggesting consistent drift detection under normal conditions with FAIR compliance. However, high-dimensional datasets posed challenges for both PCA and AE models. Update events revealed discrepancies between the two metrics, suggesting that non-linear shifts affected AE-based metrics more than PCA-based ones. Deletion events demonstrated the resilience of these metrics against data loss, but also revealed variability in the reliability of the PCA model; i.e., data drift metrics derived from PCA and AE can be effective but sensitive to certain dataset characteristics
Spelling and Grammatical Error Detection for Informal Turkish Texts with Morphologically Sensible Models
No full textTurkish is a morphologically rich language with unique characteristics such as agglutination and vowel harmony. This makes it challenging to create efficient spelling and grammatical error detection models for informal Turkish texts. Existing perspectives in the field are not enough to consider the unique characteristics of Turkish language, especially for informal texts, leading to poor precision. In this research, the project proposes to develop and discuss a morphologically sensible sequential deep learning models to aim spelling and grammatical error detection for informal Turkish texts. The proposed models are recurrent neural networks (RNN), Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), Bidirectional GRU (Bi-GRU), and Bidirectional LSTM (Bi-LSTM); the models are all types of neural network architectures, especially designed for handling sequential data. They benefit for informal Turkish-specific tasks. The models are trained and tested equal to two million rows dataset, consisting of both formal and informal Turkish texts from Turkish news, Wikipedia, and Twitter. Each of the proposed models has an accuracy of 97%. Detailed results of the 5 proposed models are presented in this paper based on classification report, confusion matrix, accuracy-loss plots, and discussion. The proposed models are highly effective to fill the void in Turkish natural language processing and improving the accuracy of spelling and grammatical error detection for informal Turkish texts. The research also checks and displays misspelled words for the put into practice informal texts with 5 text experiments, one case study for each of the proposed models, in the implementation section
Advancing Safety in Vehicles with AI-Driven Emotion Recognition
No full textEnhancing road safety in automobiles takes a thorough analysis of the driver's role, as human error is responsible for most accidents. A pivotal aspect in bolstering safety lies in the field of emotion recognition, which can effectively detect and manage emotional states to ensure a more stable driving experience. Past study limitations show the importance of developing a comprehensive system capable of identifying emotions from audio and image only. Improving road safety extends beyond outside influences to include a driver's mental well-being. The integration of Artificial Intelligence (AI) technologies in this manner creates an environment that not only promotes safety but also ensures a comfortable passenger experience. The purpose of this research is to make use of AI technology, specifically deep learning models like CNNs Long Short-term Memory (LSTM) on publicly available datasets to understand driver’s emotions through speech, text, and facial expressions. The identified emotions include happy, sad, neutral, fear, disgust, surprise, and anger. The system generates real-time alerts based on the detected emotions to enhance safety on the road. These alerts include audio, text, and visual cues to capture the driver's attention and prompt appropriate responses. To improve assurance, technology generates recommendations depending on these feelings. A music recommendation method offers songs and some quotes to help the driver's emotional state. test accuracy for image, text, and audio emotion detection is reported at 89%, 96%, and 85%, respectively
Airfare price Optimization using Quantum Computing
No full textThe aviation industry is a dynamic and complex ecosystem influenced by a variety of factors such as seat availability, distance and route length, fuel price, and other considerations, in addition to market demand and competitive dynamics. Among these features, the flight price stands out as a critical component that has a significant impact on both airlines and customers. Traditional flight price optimization strategies, while beneficial to some extent, are constrained by the inherent limitations of classical computing.
With the introduction of quantum computing, a paradigm shift in the field of optimization concerns is taking place. Quantum computing employs quantum physics ideas to do tasks previously considered to be impossible for ordinary computers. This emerging technology has the potential to change the way we tackle complex optimization issues and the ticketing industry.
The aim of this research is to look at using quantum computing techniques to optimize airline pricing tactics. Traditional methods used by airlines to compute ticket price entail extensive data analysis and simulation, with heuristic algorithms employed to navigate the vast solution spaces. With its ability to evaluate enormous amounts of data at once and exploit quantum parallelism, quantum computing is a new technology that might lead to more efficient and effective flight pricing models
Sentimental Effects of Temperature Setting After QLoRA Fine-Tuning in LLAMA 2 7B and 13B models
No full textThis research investigates the impact of temperature settings and model size on the performance of AI language models, specifically focusing on LLAMA 2 7B and 13B models post-QLoRA fine-tuning. The study is driven by the need to understand how varying conditions affect these models’ abilities to replicate known answers. The methodology includes generating data and question-answer pairs using GPT-4, fine-tuning LLAMA 2 models, and then evaluating their performance using various statistical analyses. The research’s novel contribution lies in its comprehensive approach to evaluating AI language models under different conditions. It highlights the complexities in optimizing these models and underscores the importance of considering multiple factors in their deployment and operationalization. The findings provide valuable insights for future research, particularly in exploring diverse datasets and conditions to enhance model robustness and generalizability
Exploitation of Natural Language Processing for Financial Audits
No full textFinancial audits play a crucial role in ensuring reliable corporate financial reporting. However, manually reviewing increasing volumes of textual data pose challenges to Auditors in thoroughly and consistently examining all relevant information. This research explores leveraging Natural Language Processing (NLP) to enhance the audit process. Reviewed literature identifies that NLP shows promise in areas like risk assessment, fraud detection, and workpaper review. However, most prior studies focus on limited datasets and lack integration into real audits. To address this, an NLP solution called SwiftAuditAI is designed using a large language model. The bot analyses financial statements and responds to audit questions to evaluate its ability to support audits. A comparison finds that SwiftAuditAI completes verification and extraction tasks on average 7 times faster than a human Auditor. For a sample of 20 questions, the bot's responses achieved 95% contextual accuracy compared to a human Auditor based on cosine similarity scoring of BERT embeddings