Leveraging Synthetic Data as a Tool to Combat Bias in Artificial Intelligence (AI) Model Training DOI Open Access

Jumai Adedoja Fabuyi

Journal of Engineering Research and Reports, Год журнала: 2024, Номер 26(12), С. 24 - 46

Опубликована: Ноя. 27, 2024

This study investigates the efficacy of synthetic data in mitigating bias artificial intelligence (AI) model training, focusing on demographic inclusivity and fairness. Using Generative Adversarial Networks (GANs), datasets were generated from UCI Adult Dataset, COMPAS Recidivism MIMIC-III Clinical Database. Logistic regression models trained both original to evaluate fairness metrics predictive accuracy. Fairness was assessed through parity equality opportunity, which measure balanced prediction rates equitable outcomes across groups. Fidelity diversity evaluated using statistical tests such as Kolmogorov-Smirnov (KS) Kullback-Leibler (KL) divergence, along with Inception Score, quantifies data. The results revealed significant improvements for datasets. For dataset, increased 0.72 0.89, opportunity rose 0.65 0.83, without compromising accuracy (0.82 AUC-ROC compared 0.83 data). Based findings, this research recommends employing GANs generating bias-sensitive domains enhance ensure AI models. Furthermore, integrating human-in-the-loop (HITL) systems is critical monitor address residual biases during generation. Standardized validation frameworks, including fidelity tests, should be adopted transparency consistency applications. These practices can enable organizations leverage effectively while maintaining ethical standards development deployment.

Язык: Английский

Real-Time Data Governance and Compliance in Cloud-Native Robotics Systems DOI

Onyinye Obioha-Val,

Oluwatosin Selesi-Aina,

Titilayo Modupe Kolade

и другие.

SSRN Electronic Journal, Год журнала: 2025, Номер unknown

Опубликована: Янв. 1, 2025

This study investigates the frameworks and challenges of real-time data governance compliance in cloud-native robotics systems, focusing on integrity, cloud security, regulatory adherence, cybersecurity risks. Using extensive datasets from Amazon AWS Open Data Registry, EU GDPR Enforcement Tracker, Kaggle's IoT dataset, analysis explores systems' accuracy, governance. were extracted through a standardized process: performance metrics, including latency error rates, gathered to assess system efficiency, violation records analyzed Tracker understand trends, volume metrics dataset correlated identify challenges. Together, these sources provide comprehensive insights into how systems can be optimized for realtime operations. The highlights security benefits advantages inherent frameworks, such as monitoring, automated threat detection, encryption, which collectively reduce unauthorized access risks while supporting operational efficiency. Findings indicate high accuracy (0.51% rate) low (mean 48.96 ms) across systems; however, processing time variability (standard deviation 28.61 signals need further optimization time-sensitive environments. regression violations reveals substantial penalty increase €53,789.41 per violation, emphasizing financial non-compliance. Correlation (r = 0.083 failures) suggests that external threats have greater impact than internal underscoring importance adaptive support both integrity systems.

Язык: Английский

Процитировано

0
Artificial Intelligence and Global Security: Strengthening International Cooperation and Diplomatic Relations DOI

Titilayo Modupe Kolade

SSRN Electronic Journal, Год журнала: 2024, Номер unknown

Опубликована: Янв. 1, 2024

Язык: Английский

Процитировано

0
Incorporating Privacy by Design Principles in the Modification of AI Systems in Preventing Breaches across Multiple Environments, Including Public Cloud, Private Cloud, and On-prem DOI

Samuel Ufom Okon,

Omobolaji Olufunmilayo Olateju,

Olumide Samuel Ogungbemi

и другие.

SSRN Electronic Journal, Год журнала: 2024, Номер unknown

Опубликована: Янв. 1, 2024

Язык: Английский

Процитировано

0
The Role of Artificial Intelligence (AI) in Enhancing Cybersecurity for Educational Technologies in US Public Schools DOI Open Access

Onyinye Obioha-Val

Asian Journal of Research in Computer Science, Год журнала: 2024, Номер 17(11), С. 114 - 133

Опубликована: Ноя. 25, 2024

This study investigates the role of Artificial Intelligence (AI) in enhancing cybersecurity for U.S. public schools, with primary objective evaluating AI's effectiveness reducing cyber threats and safeguarding student privacy. Specifically, assesses AI-driven security systems such as threat detection anomaly algorithms, which help schools monitor network traffic identify potential breaches real-time. Using logistic regression on data from K-12 Cybersecurity Resource Center, findings reveal that implementing AI solutions are 75% less likely to experience (p < 0.001), highlighting protective impact. Furthermore, a comparative analysis FERPA COPPA compliance reports highlights substantial reduction privacy violations among AI-using an average 0.57 per school, compared 1.50 without AI. A K-means cluster identified budget constraints (65.75%) IT staff shortages (55.25%) barriers adoption. To address these obstacles, recommends phased technology upgrades increased funding workforce training critical strategies facilitate integration enhance across educational institutions. These strategic interventions essential optimizing systems, making it feasible resource-constrained adopt maintain advanced measures. The study’s contribute growing body knowledge provide actionable insights policymakers administrators seeking strengthen protection school environments.

Язык: Английский

Процитировано

0
Leveraging Synthetic Data as a Tool to Combat Bias in Artificial Intelligence (AI) Model Training DOI Open Access

Jumai Adedoja Fabuyi

Journal of Engineering Research and Reports, Год журнала: 2024, Номер 26(12), С. 24 - 46

Опубликована: Ноя. 27, 2024

This study investigates the efficacy of synthetic data in mitigating bias artificial intelligence (AI) model training, focusing on demographic inclusivity and fairness. Using Generative Adversarial Networks (GANs), datasets were generated from UCI Adult Dataset, COMPAS Recidivism MIMIC-III Clinical Database. Logistic regression models trained both original to evaluate fairness metrics predictive accuracy. Fairness was assessed through parity equality opportunity, which measure balanced prediction rates equitable outcomes across groups. Fidelity diversity evaluated using statistical tests such as Kolmogorov-Smirnov (KS) Kullback-Leibler (KL) divergence, along with Inception Score, quantifies data. The results revealed significant improvements for datasets. For dataset, increased 0.72 0.89, opportunity rose 0.65 0.83, without compromising accuracy (0.82 AUC-ROC compared 0.83 data). Based findings, this research recommends employing GANs generating bias-sensitive domains enhance ensure AI models. Furthermore, integrating human-in-the-loop (HITL) systems is critical monitor address residual biases during generation. Standardized validation frameworks, including fidelity tests, should be adopted transparency consistency applications. These practices can enable organizations leverage effectively while maintaining ethical standards development deployment.

Язык: Английский

Процитировано

0