Machine Learning Intelligence Research Articles

1325-P: Applications of Artificial Intelligence–Machine Learning by Using Spectral Analysis of Arterial Pulse to Discriminate the Vascular Effects of Metabolic Syndrome and Prediabetes

1325-P: Applications of Artificial Intelligence–Machine Learning by Using Spectral Analysis of Arterial Pulse to Discriminate the Vascular Effects of Metabolic Syndrome and Prediabetes

A Study of the Impact of Digitalization in HR in Private Organization

In the past human resources management mainly focused on tasks such, as keeping track of employees handling payments and managing benefits. However, there has been a shift towards planning. Aligning HR practices with organizational goals. This shift emphasizes talent acquisition, employee development and performance management. Employee management is crucial for any organization as it involves supervising and coordinating the workforce to achieve objectives. With the advancements in technology over the year’s employee management has evolved through human resource management. Digital HRM goes beyond technology utilization it integrates “cutting edge technologies like cloud computing, artificial intelligence (AI) machine learning (ML)” analytics tools and social media platforms into core HR goals. This study delves into how the implementation of technologies has impacted the way resources (HR) functions are managed at private organizations, in Oman. It seeks to uncover the changes brought about by digitalization in HR processes like recruitment, onboarding and employee record keeping. The research also explores the reasons behind organization’s adoption of tools in their HR operations and how the level of readiness influences the success of their digital HR endeavors.

Harnessing adversarial machine learning for advanced threat detection: AI-driven strategies in cybersecurity risk assessment and fraud prevention

The abstract is "The rapid evolution of cyber threats necessitates innovative defenses, particularly in the domains of risk assessment and fraud detection. This paper explores the integration of Artificial Intelligence (AI) and Adversarial Machine Learning (ML) techniques as a formidable strategy against increasingly sophisticated cyber-attacks. We present a comprehensive framework that leverages AI to dynamically assess cybersecurity risks and detect fraudulent activities with unprecedented accuracy and speed. Firstly, we delve into the foundational principles of adversarial machine learning, outlining how these techniques can be employed to simulate potential cyber threats, thereby enabling the development of more resilient AI-driven cybersecurity systems. We highlight the dual role of adversarial ML in both enhancing security defenses and potentially serving as a vector for sophisticated attacks, underscoring the importance of developing robust, adversarial-resistant models. Subsequently, we introduce a novel adaptive risk assessment methodology that incorporates real-time data analysis, machine learning algorithms, and predictive modeling to accurately identify and prioritize threats. This method adapts to the evolving digital landscape, ensuring that cybersecurity measures are always one step ahead of potential attackers. In the context of fraud detection, we explore -how AI algorithms can sift through vast datasets to detect anomalies and patterns indicative of fraudulent behavior. Through case studies and empirical analysis, we demonstrate the effectiveness of AI in identifying fraud across various sectors, from financial transactions to online identity verification processes. Our research contributes to the cybersecurity field by providing a detailed examination of how AI and adversarial ML can be harnessed to fortify digital defenses, improve risk assessment techniques, and enhance fraud detection capabilities. The insights garnered from this study not only advance theoretical understanding but also offer practical guidance for organizations seeking to implement AI-driven security solutions. As cyber threats continue to evolve, the integration of AI and adversarial ML in cybersecurity strategies will be paramount in safeguarding digital assets and maintaining the integrity of online systems."

Towards real-time earthquake forecasting in Chile: Integrating intelligent technologies and machine learning

This paper proposes an innovative approach towards real-time earthquake forecasting in Chile by integrating intelligent technologies and machine learning methods. Earthquakes pose significant risks to communities and infrastructure in Chile, making accurate and timely forecasting crucial for disaster preparedness and mitigation. Traditional forecasting methods have limitations in providing real-time insights into seismic activity. In contrast, intelligent technologies such as artificial intelligence (AI) and machine learning offer promising avenues for enhancing prediction accuracy and speed. A new earthquake forecasting method using a modified clustering approach LMSCAN(Local Maxima-based Spatio-Cluster Analysis Network) and enhanced neural network time series analysis LSTM-IC (Long Short Term Memory Inverse Correlation) is presented in this paper. This neural network-based technology has been utilized to forecast earthquakes in the Chile region, one of the countries with the largest seismic activity. This study explores the integration of intelligent systems with machine learning algorithms to analyze seismic data and predict earthquake occurrences in Chile. By leveraging historical seismic data, sensor networks, and advanced predictive models, our approach aims to provide timely warnings and insights into seismic events, thereby improving disaster response and resilience. The proposed framework holds the potential to revolutionize earthquake forecasting by enabling real-time monitoring and proactive measures to safeguard communities and infrastructure in Chile and beyond. The remarkable 95 % accuracy achieved by this model is a testament to its exceptional learning process, which sets it apart from other models. Its ability to learn and adapt to new data is unparalleled, allowing it to forecast incredibly precisely and produce highly reliable results.

A methodological integration of fisher score technique with intelligent machine learning methods for ball bearing fault investigation

This paper presents an efficient approach to detect, diagnose and estimate the severity of failures in various components of bearings in induction motors using vibration signature analysis. This automated method integrates the Fisher Score feature selection method and an efficient hyperparameter tuning model with machine learning models, including Support Vector Machine (SVM), k-Nearest Neighbor (k-NN) and Decision Tree (DT), to accurately classify defects in bearings. This approach ensures accurate classification of bearing defects within less computation time. This work is carried out with vibration signals, recorded from a laboratory experimental setup on Machine Fault Simulator (MFS), focusing on ball bearing with defects in inner race, outer race and ball itself. Time and Frequency domain analysis are employed to compute the features for fault investigation in ball bearings using machine learning models. The computed results demonstrate that the proposed feature selection method with hyperparameter tuning achieved remarkable maximum accuracy among X, Y and XY combinations of datasets, with 97% in DT, 94% in SVM and 95.23% in k-NN models during the frequency domain analysis. Notably, these model accuracies improved to 99.04% in DT, 98% in SVM and 98% in k-NN during further analysis with Fisher Score technique. Consequently, the testing loss using this hyperparameter tuning function remains very low. Overall, this paper compares the results of time and frequency domain analysis and introduces a promising approach for both efficient and accurate fault detection and severity estimation in bearings of induction motors, potentially reducing the need for extensive manpower and sensor usage.

Waste-to-energy poly-generation scheme for hydrogen/freshwater/power/oxygen/heating capacity production; optimized by regression machine learning algorithms

Utilization of machine learning techniques in the analysis and enhancement of poly-generation energy systems improves their efficiency and sustainability. Also, waste-to-energy systems propose a hopeful answer for both waste management and sustainable energy and water production. The production of hydrogen and freshwater through these systems not only provides valuable resources but also contributes to environmental sustainability. This research utilizes artificial intelligence's machine learning algorithms to examine and enhance a waste-to-energy system within a poly-generation energy system that generates hydrogen, freshwater, power, oxygen, and hot air. The integrated system consists of a waste combustion chamber, a proton exchange membrane electrolyzer, a supercritical carbon dioxide Brayton cycle, and a desalination system. Based on the findings, the machine learning algorithms exhibiting R-squared values exceeding 99% are considered strong fits. Additionally, all algorithms with high predicted R-squared values have the capability to accurately forecast new data using the provided training data. The emissions of 669.7 g/kWh and an efficiency of 69.62% are achieved when the pressure ratio is 10.73 and the temperature is 863.6 °C under optimal conditions. The accuracy and validity of the machine learning techniques are further confirmed by the strong agreement with thermodynamic modeling.

UTILIZAREA IMAGINILOR SATELITARE ÎN AGRICULTURĂ. STUDIU DE CAZ

All sizes of farms can benefit from satellite imagery, not only big producers. When paired with artificial intelligence (AI) and deep machine learning techniques, satellite photography becomes an effective tool for monitoring agricultural conditions and anticipating issues in the field. As a result, using satellite photos to guide crop farming choices can help determine when to apply nutrients and irrigation. This paper focuses on monitoring through satellite sensors with an emphasis on the facilities offered by the European Copernicus Program through Sentinel-2 satellites the crops from a farm from Calarasi County, Borcea commune.

Hybrid intelligent machine learning system for modeling phosphate ore processing processes

Hybrid intelligent machine learning system for modeling phosphate ore processing processes

Identifying necrotizing soft tissue infection using infectious fluid analysis and clinical parameters based on machine learning algorithms

BackgroundDetermining the presence of necrotizing soft tissue infection (NSTI) poses a significant hurdle. As of late, there has been a notable increase in the application of artificial intelligence (AI) machine learning techniques in identifying diseases, a shift that can be attributed to their exceptional efficiency, unbiased nature, and high precision. MethodsInformation was gathered from a cohort of 13 patients suffering from NSTI, alongside 12 patients with cellulitis. The construction of NSTI diagnostic machine learning models utilized four different algorithms, specifically random forest, k-nearest neighbors (KNN), support vector machine (SVM), and logistic regression. These models were constructed based on 28 distinctive attributes identified through statistical examination. Following this, the diagnostic efficiency of each algorithms was evaluated. A novel random forest model, streamlined for clinical use, was later developed by focusing on 6 attributes that had the most pronounced influence on the accuracy of our initial random forest model. ResultsThe following data was noted regarding the sensitivity and specificity of the four NSTI diagnostic models:logistic regression displayed 78.2 % and 83.7 %, KNN presented 79.1 % and 87.1 %, SVM showed 83.5 % and 86.3 %, and random forest exhibited 89.6 % and 92.9 %, respectively. In comparison, lactate levels in fluid demonstrated 100 % sensitivity and 76.9 % specificity at an optimal cut-off point of 69.6 mg/dL. Among all four machine learning models, random forest outperformed the others and also showed better results than fluid lactate. A newly constructed random forest model, created using 6 of the 13 identified features, displayed promising results in diagnosing NSTI, having a sensitivity and specificity of 90.2 % and 92.2 %, respectively. ConclusionsDeveloping a diagnostic model for NSTI employing the random forest algorithm has resulted in a diagnostic technique that is more efficient, cost-effective, and expedient. This approach could provide healthcare practitioners with the tools to identify and manage NSTI with greater efficacy.

Computer-aided urban energy systems cyber attach detection and mitigation: Intelligence hybrid machine learning technique for security enhancement of smart cities

This paper introduces a sophisticated computer-aided system for enhancing the security of smart cities through urban cyber attack detection and mitigation. The proposed system employs a comprehensive mathematical framework that models the dynamics of cyber attacks, detection, and mitigation, capturing the intricate relationships within a smart city environment. Leveraging a hybrid approach combining Transfer Learning and Meta-Learning, the system adapts to evolving threats and tasks, showcasing superior performance in comparison to standalone methods. Simulation results demonstrate the system’s robustness, with low False Positive and False Negative Rates, high System Security, and effective Mitigation Strategies. This research contributes a state-of-the-art solution for securing smart cities, leveraging intelligent hybrid machine learning techniques.

Triboelectric sensor-empowered intelligent mouse combined with machine learning technology strides toward a computer security system

The computer mouse is one of the most common tools used for human-machine interfacing. The human behavior studying by click operation using an intelligent mouse and related security identification of computers remains a great challenge. Here, we develop an intelligent mouse embedded with a contact-separation adaptable triboelectric sensor based on the intrinsic mouse button click operation. With the unique characteristics of different individuals who use the intelligent mouse and machine learning algorithms, an identity recognition system was designed. Through processes such as signal processing, feature extraction, and model training, high-accuracy (98.4 %) identification of different individuals based on the signals from the embedded triboelectric sensor in the mouse was achieved. In addition, we have developed a security alarm system for detecting unauthorized users operating the computer, and successfully validated the practicality of this system. A system for monitoring user behaviors while operating the computer has been designed to meet the monitoring needs of user computer usage states. This system can monitor users' gaming, working, or studying states, while effectively protecting users' privacy. In general, this work reveals a new strategy for security identification of computers which can be potentially applied in cyber and information security.

Methodology for forming a digital farming management system

The paper presents methodological approaches for the creation of a digital farming management system (DFMS). A convergent approach, based on cognitive (conceptual) analysis methods, is employed in the research and applied to the formation of adaptive landscape farming systems. The fundamental principles of organizing DFMS include crop and environmental monitoring (in situ, remote sensing); the formation of farming system archetypes based on the analysis of long-term field experiments; spatial object modeling and land typology using GIS; planning and support for agrotechnologies to adapt to natural and economic conditions; modeling ecosystem services and biodiversity; assessing the impact on the sustainability and economics of crop production. The system is implemented using geoinformation models in a specific geographic coordinate. DFMS involves conducting a "inventory" of natural and production resources, as well as identifying limits of climatic, soil, and agrolandscape parameters at different levels of land use intensity. At each stage of organizing system blocks, methods of intelligent data analysis and machine learning are used, with the core of the system relying on the use of knowledge bases and logical rules of the subject area. A key element of the system is the scaling of the results of long-term field experiments and accumulated knowledge in different management areas based on the parameterization of the multi-level variability of farming systems and the formation of their archetypes. The practical implementation of the main provisions of DFMS allows approaching the solution of key issues related to reducing the level of uncertainty and associated risks in agriculture. This is achieved through scientifically justified organization of rational land use, increasing the resilience of crop production in different land use conditions, and providing information support to rural producers.

The Impact of Business Continuity on Supply Chain Practices and Resilience Due to COVID-19

Background: Business continuity entails the potential negative consequences of uncertainty on a firm’s ability to achieve strategic objectives. The COVID-19 pandemic significantly impacted business continuity due to lockdowns, travel restrictions, and social distancing measures. Consequently, firms adopted specific supply chain (SC) practices to effectively navigate this global crisis. Methods: This research adopted a stochastic approach based on Bayesian Networks to evaluate the implications of business continuity on firms’ decisions to embrace SC practices, focusing on omnichannel strategies, SC coordination, and technologies such as artificial intelligence systems, big data and machine learning, and mobile applications. Results: Our findings revealed that firms facing disruption in a single performance area can apply specific strategies to maintain resilience. However, multiple areas of underperformance necessitate a varied approach. Conclusions: According to our empirical analysis, omnichannel strategies are critical when disruptions simultaneously impact quality, inventory, sales, and ROI, particularly during major disruptions such as the COVID-19 pandemic. AI and big data become vital when multiple risks coalesce, enhancing areas such as customer service and supply chain visibility. Moreover, supply chain coordination and mobile app adoption are effective against individual performance risks, proving crucial in mitigating disruption impacts across various business aspects. These findings help policy-makers and business owners to have a better understanding of how business continuity based on performance resistance to disruptions pushes companies to adopt different practices including new technologies and supply chain coordination. Accordingly, they can use the outputs of this study to devise strategies for improving resilience considering their supply chain vulnerabilities.

An Improved Intelligent Machine Learning Approach to Music Recommendation Based on Big Data Techniques and DSO Algorithms

INTRODUCTION: In an effort to enhance the quality of user experience in using music services and improve the efficiency of music recommendation platforms, researching accurate and efficient music recommendation methods and constructing an accurate real-time online recommendation platform are the key points for the success of a high-quality music website platform.OBJECTIVES: To address the problems of incomplete signal feature capture, insufficient classification efficiency and poor generalization of current music recommendation methods.METHODS: Improve the deep confidence network to construct music recommendation algorithm by using big data and intelligent optimization algorithm. Firstly, music features are extracted by analyzing the principle of music recommendation algorithm, and evaluation indexes of music recommendation algorithm are proposed at the same time; then, combined with the deep sleep optimization algorithm, a music recommendation method based on improved deep confidence network is proposed; finally, the efficiency of the proposed method is verified through the analysis of simulation experiments.RESULTS: While meeting the real-time requirements, the proposed method improves the music recommendation accuracy, recall, and coverage.CONCLUSION: Solves the questions of incomplete signal feature capture, insufficient classification efficiency, and poor generalization of current music recommendation algorithms.

Abstract LB122: JNJ-87890387, a novel ENPP3 bispecific T-cell redirector (ENPP3xCD3) with tumor selectivity through targeting apical surface antigens

Abstract Clinical success of T-cell redirection therapy for solid tumors has been limited potentially due to a lack of target specificity against cancer, leading to on-target off-tumor toxicity at sub-efficacious doses. ENPP3 (ectonucleotide pyrophosphatase/phosphodiesterase family member 3) is a transmembrane protein that was identified with the help of an artificial intelligence machine learning algorithm. ENPP3 is characterized by apically restricted expression in most normal tissues and elevated, depolarized expression in several solid tumors, thus potentially providing tumor selectivity. ENPP3 expression exhibits a high prevalence in several solid tumors including renal cell carcinoma (RCC) - clear cell (93%) and papillary (78%), lung adenocarcinoma (50%), endometrioid uterine (53%) and ovarian (47%) cancers, colorectal carcinoma (51%), and hepatocellular carcinoma (HCC, 47%). JNJ-87890387 is a fully human bispecific antibody with high affinity and specific binding to ENPP3 on tumor cells and lower affinity binding to cluster of differentiation (CD)3 on T-cells. In vitro, JNJ-87890387 demonstrated potent (pM) T-cell activation and cytotoxicity against tumor cell lines with different levels of endogenous ENPP3 membrane expression, and absence of killing against cells lacking ENPP3 expression, confirming JNJ-87890387 specificity. In vivo, JNJ-87890387 demonstrated ENPP3-expression-dependent potent anti-tumor activity in multiple cell line-derived and patient tumor-derived RCC and HCC xenograft models with complete tumor regressions, and evidence of dose-dependent CD4+ and CD8+ T cell infiltration into the tumors. Preclinical safety assessment in the cynomolgus monkey with a tool molecule using the therapeutic ENPP3 binding arm but with a cynomolgus CD3 cross-reactive arm only identified toxicities consistent with the modality. In summary, these data support JNJ-87890387’s clinical development with an ongoing first-in-human Phase I study to evaluate the safety and preliminary anti-tumor activity in advanced-stage solid tumors with a high prevalence of ENPP3 expression. Citation Format: Smruthi Vijayaraghavan, Pankaj Seth, Annmarie Winkis, Kristen Chevalier, Alexa Marthaler, Katrin Sproesser, Vince Torti, Nirav Shah, Eilyn Lacy, Anna-Lena Frisk, Heather Deutsch, Gerald Chu, Michael Sharp, Jonathan Pabalan, Cholpon Tilegenova, Katherine Thorton, Josh Lauring, Bethany Mattson, Ken Tian, Jacqueline Kinyamu-Akunda, Laurie Lenox, Wan Cheung Cheung, Sheri Moores, Sylvie Laquerre. JNJ-87890387, a novel ENPP3 bispecific T-cell redirector (ENPP3xCD3) with tumor selectivity through targeting apical surface antigens [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(7_Suppl):Abstract nr LB122.

A Machine Learning Algorithm Improves the Diagnostic Accuracy of the Histologic Component of Antibody Mediated Rejection (AMR-H) in Cardiac Transplant Endomyocardial Biopsies.

BackgroundPathologic antibody mediated rejection (pAMR) remains a major driver of graft failure in cardiac transplant patients. The endomyocardial biopsy remains the primary diagnostic tool but presents with challenges, particularly in distinguishing the histologic component (pAMR-H) defined by 1) intravascular macrophage accumulation in capillaries and 2) activated endothelial cells that expand the cytoplasm to narrow or occlude the vascular lumen. Frequently, pAMR-H is difficult to distinguish from acute cellular rejection (ACR) and healing injury. With the advent of digital slide scanning and advances in machine deep learning, artificial intelligence technology is widely under investigation in the areas of oncologic pathology, but in its infancy in transplant pathology. For the first time, we determined if a machine learning algorithm could distinguish pAMR-H from normal myocardium, healing injury and ACR. Materials and MethodsA total of 4,212 annotations (1,053 regions of normal, 1,053 pAMR-H, 1,053 healing injury and 1,053 ACR) were completed from 300 hematoxylin and eosin slides scanned using a Leica Aperio GT450 digital whole slide scanner at 40X magnification. All regions of pAMR-H were annotated from patients confirmed with a previous diagnosis of pAMR2 (>50% positive C4d immunofluorescence and/or >10% CD68 positive intravascular macrophages). Annotations were imported into a Python 3.7 development environment using the OpenSlide™ package and a convolutional neural network approach utilizing transfer learning was performed. ResultsThe machine learning algorithm showed 98% overall validation accuracy and pAMR-H was correctly distinguished from specific categories with the following accuracies: normal myocardium (99.2%), healing injury (99.5%) and ACR (99.5%). ConclusionOur novel deep learning algorithm can reach acceptable, and possibly surpass, performance of current diagnostic standards of identifying pAMR-H. Such a tool may serve as an adjunct diagnostic aid for improving the pathologist's accuracy and reproducibility, especially in difficult cases with high inter-observer variability. This is one of the first studies that provides evidence that an artificial intelligence machine learning algorithm can be trained and validated to diagnose pAMR-H in cardiac transplant patients. Ongoing studies include multi-institutional verification testing to ensure generalizability.

인공지능 기반의 자동화된 통합보안관제시스템 모델 연구

In today's growing threat environment, rapid and effective detection and response to security events is essential. To solve these problems, many companies and organizations respond to security threats by introducing security control systems. However, existing security control systems are experiencing difficulties due to the complexity and diverse characteristics of security events. In this study, we propose an automated integrated security control system model based on artificial intelligence. It is based on deep learning, an artificial intelligence technology, and provides effective detection and processing functions for various security events. To this end, the model applies various artificial intelligence algorithms and machine learning methods to overcome the limitations of existing security control systems. The proposed model reduces the operator's workload, ensures efficient operation, and supports rapid response to security threats.

Abstract 4915: Identification and validation of novel prognostic genetic markers in HER2-negative advanced gastric cancer (AGC) by artificial intelligence (AI) deep learning and machine learning algorithm

Abstract Introduction: Despite advances in anticancer treatments, there has been urgent need to identify novel prognostic markers for HER2-negative AGC. We propose AI based deep learning and machine learning pipeline to explore genetic markers that predict patient response to chemotherapy in HER2-negative AGC. Methods: We retrospectively assessed 179 stage IV HER2-negative AGC patients who have been treated with first-line chemotherapy in Yonsei Cancer Center, Korea between 2015 and 2021. In-house targeted sequencing panel data (CancerSCANTM and CancerMaster) were used to identify candidate prognostic markers. DeepSurv was mainly used to analyze the progression-free survival (PFS), which is a deep learning algorithm that can investigate prognostic roles. The SHapley Additive exPlanations (SHAP) method was applied to open the deep learning models’ black-box and calculate the importance ranking of gene features. Machine learning models-random survival forest, elastic-net, and lasso-were also investigated to identify potentially meaningful gene variants. To rank the gene features, we calculated variable importance (VIMP) for random survival forest, and penalized regression coefficients for elastic-net and lasso. All deep learning and machine learning models were trained using 5-folds cross-validation in 1,000 iterations of re-sampling bootstrapping with hyperparameter tuning by grid search. To classify responder or non-responder related genetic markers to chemotherapy, we used the signs of averaged penalize coefficients obtained from elastic-net and lasso. Based on the median linear predictor with candidate gene features, patients were classified into high-risk and low-risk group. Results: A total of 7,824 common variants were analyzed and 20 prognostic genetic markers were identified based on the top average values of SHAP, VIMP, and penalized regression coefficients. The top 10 non-responder genetic markers are included as follows: PARP4(p.S873N), CHUK(p.V268I), BARD1(359_365del), CREBBP(p.Y1125F), BARD1(p.P24S), PHLPP2(p.R1312Q), FAT3(p.Q3375R), CASP5(p.T106A), PHLPP2(p.I544V), and PTCH2(p.T988M). Otherwise, the top 10 responder genetic markers are defined as follows: IL7R(p.I66T), NOTCH4(L16delinsLL), FGFR3, BCL2A1(p.G82D), ZNF217(p.T548I), BRCA1(p.K1183R), FANCA(p.A412V), ADGRA2, CDKN2B, and FANCM(p.S175F). The high-risk group had worse PFS compared with low-risk group (median PFS 4.1 vs. 9.1 months; P-value <0.0001). Conclusion: This study showed the potential of AI deep learning and machine learning pipeline that employs an innovate prognostic genetic markers for HER2-negative AGC. Current ongoing analyses of larger cohort with more comprehensive clinical data will be presented. Citation Format: Sejung Park, Seok-Jae Heo, Choong-Kun Lee, Yaeji Lee, Woo Sun Kwon, Jingmin Che, Minseok Kim, Hyun Cheol Chung, Sun Young Rha. Identification and validation of novel prognostic genetic markers in HER2-negative advanced gastric cancer (AGC) by artificial intelligence (AI) deep learning and machine learning algorithm [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4915.

PersistVerify: Federated model ownership verification with spatial attention and boundary sampling

Federated learning, known for its emphasis on privacy and resource efficiency, has emerged as a transformative paradigm in the fields of artificial intelligence and industrial machine learning. However, with the advancement and widespread adoption of this technology, the concern of model theft in federated learning models becomes increasingly prominent. While current endeavors concentrate on protecting individual deep learning models, limited attention has been dedicated to those involved in federated learning, especially within decentralized environments. To fill this void, we introduce PersistVerify, an innovative approach for verifying model copyright within the federated learning framework. PersistVerify integrates boundary sample selection and spatial attention mechanisms, enhancing the robustness and confidentiality of neural network backdoor watermarks to ensure secure verification of model ownership. PersistVerify constructs a confidential and robust watermark dataset through three sequential steps: latent representation extraction, boundary sample selection, and spatial attention-based trigger embedding. Our research addresses the pervasive challenge of reliable model ownership verification in federated learning, demonstrating the efficacy of PersistVerify. This study provides valuable insights and methodologies to maintain the confidentiality and reliability of neural network backdoor watermarks in the domains of artificial intelligence and federated learning.

A study of machine learning object detection performance for phased array ultrasonic testing of carbon fibre reinforced plastics

The growing adoption of Carbon Fibre Reinforced Plastics (CFRPs) in the aerospace industry has resulted in a significant reliance on Non-Destructive Evaluation (NDE) to ensure the quality and integrity of these materials. The interpretation of large amounts of data acquired from automated robotic ultrasonic scanning by expert operators is often time consuming, tedious, and prone to human error creating a bottleneck in the manufacturing process. However, with ever growing trend of computing power and digitally stored NDE data, intelligent Machine Learning (ML) algorithms have been gaining more traction than before for NDE data analysis. In this study, the performance of ML object detection models, statistical methods for defect detection, and traditional amplitude thresholding approaches for defect detection in CFRPs were compared. A novel augmentation technique was used to enhance synthetically generated datasets used for ML model training. All approaches were tested on real data obtained from an experimental setup mimicking industrial conditions, with ML models showing improvement over amplitude thresholding and statistical thresholding techniques. The advantages and limitations of all methods are reported and discussed.