7 research outputs found
A Comparative Analysis of Statistical Modeling and Machine Learning Techniques for Predicting the Lifetime of Light Emitting Diodes From Accelerated Life Testing
No full textThis study re-examines the failure behavior of high-brightness blue LEDs under accelerated life testing using multivariable stress models. Current, temperature, RH, and their interactions are analyzed, revealing that the lognormal distribution fits lifetime data better than the Weibull. The Intel model outperforms the Peck model for stress-life relationships. Machine learning is also explored for lifetime prediction, though its limitations under normal conditions are noted.This work uses multivariable life stress models to revisit the catastrophic failure of high-brightness blue light emitting diodes (LEDs) under accelerated life testing (ALT). The stress factors, current, temperature, relative humidity (RH), and their interactions are considered in lifetime studies. First, we show that the lognormal distribution fits the experimental data much better than the Weibull distribution using the standard Kolmogorov-Smirnov test. Furthermore, the best life-stress relationship is the Intel model rather than the peck model used by Nogueira et al. (2016). Additionally, based on the accelerated data, machine learning (ML) techniques are employed to predict the lifetime of LEDs under normal operating conditions. However, the study highlights the limitations of ML in accurately predicting lifetime
Encryption-based Adversarial Defense for Resilient Facial Recognition Deep Learning Models
No full textacial recognition is a vital technology that has been widely adopted for various
applications, including security and identification. However, facial recognition models
are vulnerable to adversarial attacks, raising concerns about their security and
reliability. This project investigates the impact of encryption on the security and
robustness of facial recognition deep learning models (FR-DL) against adversarial
attacks.
A comprehensive literature review was conducted to identify research gaps, explore
defense methods, evaluate datasets, and examine model accuracies. Based on our
literature review, image transformation (encryption) has been identified as a de-
fense method that oers a high level of reliability, feasible implementation, and
demonstrated accuracy. This study will investigate weather deep learning models
can eectively learn from images that have been encrypted, and how can encryption
improve their robustness against adversarial attacks.
The proposed methodology involves data collection and processing to curate a suitable
dataset. Leveraging the expansive and diverse VGGFace2 dataset, we will train and
test deep learning models. Pixel shuing will be applied to the dataset as the encryp-
tion method. The resultant encrypted data will serve as the foundation for building
and training the models. Rigorous testing will assess the models’ resilience against
adversarial attacks. Continuous performance analysis and accuracy assessments will be integral, aiming to achieve a 90% accuracy rate throughout the process.
The expected outcome of the project is to provide valuable insight into how en-
cryption impacts the robustness of deep learning models against adversarial attacks,
contributing to the development of secure AI systems. This aligns with Vision 2030,
transforming Saudi Arabia into a modern, economically, and socially vibrant nation
Data mining an NLP
No full textThe rapid advancements in artificial intelligence (AI) have primarily focused on the process of learning from data to acquire knowledge for smart systems. However, the concept of machine unlearning has emerged as a transformative paradigm shift in the field of AI, due to the amount of false information that have been learned over the past. Machine unlearning refers to the ability of AI systems to reverse or discard previously acquired knowledge or patterns, enabling them to adapt and refine their understanding in response to changing circumstances or new insights. This paper explores the concept of machine unlearning, its implications, methods, challenges, and potential applications. The paper begins by providing an overview of the traditional learning-based approaches in AI and the limitations they impose on system adaptability and agility. It then delves into the concept of machine unlearning, discussing various techniques and algorithms employed to remove or modify learned knowledge from AI models or datasets.Effat Universit
The Effectiveness of Data Visualization Techniques for Communicating Insights: A Cross-domain Analysis
No full textSARS-CoV-2 infection and venous thromboembolism after surgery: an international prospective cohort study
No full textSARS-CoV-2 has been associated with an increased rate of venous thromboembolism in critically ill patients. Since surgical patients are already at higher risk of venous thromboembolism than general populations, this study aimed to determine if patients with peri-operative or prior SARS-CoV-2 were at further increased risk of venous thromboembolism. We conducted a planned sub-study and analysis from an international, multicentre, prospective cohort study of elective and emergency patients undergoing surgery during October 2020. Patients from all surgical specialties were included. The primary outcome measure was venous thromboembolism (pulmonary embolism or deep vein thrombosis) within 30 days of surgery. SARS-CoV-2 diagnosis was defined as peri-operative (7 days before to 30 days after surgery); recent (1–6 weeks before surgery); previous (≥7 weeks before surgery); or none. Information on prophylaxis regimens or pre-operative anti-coagulation for baseline comorbidities was not available. Postoperative venous thromboembolism rate was 0.5% (666/123,591) in patients without SARS-CoV-2; 2.2% (50/2317) in patients with peri-operative SARS-CoV-2; 1.6% (15/953) in patients with recent SARS-CoV-2; and 1.0% (11/1148) in patients with previous SARS-CoV-2. After adjustment for confounding factors, patients with peri-operative (adjusted odds ratio 1.5 (95%CI 1.1–2.0)) and recent SARS-CoV-2 (1.9 (95%CI 1.2–3.3)) remained at higher risk of venous thromboembolism, with a borderline finding in previous SARS-CoV-2 (1.7 (95%CI 0.9–3.0)). Overall, venous thromboembolism was independently associated with 30-day mortality (5.4 (95%CI 4.3–6.7)). In patients with SARS-CoV-2, mortality without venous thromboembolism was 7.4% (319/4342) and with venous thromboembolism was 40.8% (31/76). Patients undergoing surgery with peri-operative or recent SARS-CoV-2 appear to be at increased risk of postoperative venous thromboembolism compared with patients with no history of SARS-CoV-2 infection. Optimal venous thromboembolism prophylaxis and treatment are unknown in this cohort of patients, and these data should be interpreted accordingly
Pancreatic surgery outcomes: multicentre prospective snapshot study in 67 countries
No full textBackground: Pancreatic surgery remains associated with high morbidity rates. Although postoperative mortality appears to have
improved with specialization, the outcomes reported in the literature reflect the activity of highly specialized centres. The aim of
this study was to evaluate the outcomes following pancreatic surgery worldwide.
Methods: This was an international, prospective, multicentre, cross-sectional snapshot study of consecutive patients undergoing
pancreatic operations worldwide in a 3-month interval in 2021. The primary outcome was postoperative mortality within 90 days of
surgery. Multivariable logistic regression was used to explore relationships with Human Development Index (HDI) and other parameters.
Results: A total of 4223 patients from 67 countries were analysed. A complication of any severity was detected in 68.7 per cent of
patients (2901 of 4223). Major complication rates (Clavien–Dindo grade at least IIIa) were 24, 18, and 27 per cent, and mortality rates
were 10, 5, and 5 per cent in low-to-middle-, high-, and very high-HDI countries respectively. The 90-day postoperative mortality
rate was 5.4 per cent (229 of 4223) overall, but was significantly higher in the low-to-middle-HDI group (adjusted OR 2.88, 95 per
cent c.i. 1.80 to 4.48). The overall failure-to-rescue rate was 21 per cent; however, it was 41 per cent in low-to-middle- compared
with 19 per cent in very high-HDI countries.
Conclusion: Excess mortality in low-to-middle-HDI countries could be attributable to failure to rescue of patients from severe
complications. The authors call for a collaborative response from international and regional associations of pancreatic surgeons to
address management related to death from postoperative complications to tackle the global disparities in the outcomes of pancreatic
surgery (NCT04652271; ISRCTN95140761)
SARS-CoV-2 vaccination modelling for safe surgery to save lives: data from an international prospective cohort study
No full textBackground: Preoperative SARS-CoV-2 vaccination could support safer elective surgery. Vaccine numbers are limited so this study aimed to inform their prioritization by modelling.
Methods: The primary outcome was the number needed to vaccinate (NNV) to prevent one COVID-19-related death in 1 year. NNVs were based on postoperative SARS-CoV-2 rates and mortality in an international cohort study (surgical patients), and community SARS-CoV-2 incidence and case fatality data (general population). NNV estimates were stratified by age (18-49, 50-69, 70 or more years) and type of surgery. Best- and worst-case scenarios were used to describe uncertainty.
Results: NNVs were more favourable in surgical patients than the general population. The most favourable NNVs were in patients aged 70 years or more needing cancer surgery (351; best case 196, worst case 816) or non-cancer surgery (733; best case 407, worst case 1664). Both exceeded the NNV in the general population (1840; best case 1196, worst case 3066). NNVs for surgical patients remained favourable at a range of SARS-CoV-2 incidence rates in sensitivity analysis modelling. Globally, prioritizing preoperative vaccination of patients needing elective surgery ahead of the general population could prevent an additional 58 687 (best case 115 007, worst case 20 177) COVID-19-related deaths in 1 year.
Conclusion: As global roll out of SARS-CoV-2 vaccination proceeds, patients needing elective surgery should be prioritized ahead of the general population
