Prior Approaches Research Articles

An efficient sparse code shrinkage technique for ECG denoising using empirical mode decomposition

Accurate denoising of Electrocardiogram (ECG) signals is essential for reliable cardiac diagnostics, but traditional methods often struggle with high-frequency noise and artifacts, leading to potential misinterpretations. It is often impeded by interference such as power line interference (PLI) and Gaussian noise. To address this challenge, we suggest a novel ECG denoising technique that combines empirical mode decomposition (EMD) with wavelet domain sparse code shrinking. Our approach first decomposes the noisy ECG signal into Intrinsic Mode Functions (IMFs) using EMD. These IMFs are then transformed into the wavelet domain, where a sparse code shrinking function is applied to effectively reduce both Gaussian noise and PLI while preserving the integrity of the original signal. The effectiveness of the technique is assessed on the MIT-BIH database, where it shows marked improvements in Signal-to-Noise Ratio (SNR), Mean Squared Error (MSE), and Percentage Root Mean Square Difference (PRD). The suggested approach demonstrates improved SNR and reduced MSE when compared to prior approaches, which suggests that the ECG signals are clearer and more precise. This method presents a rather effective approach to enhancing ECG analysis as it is important for diagnosis and interpretation. At 10 dB SNR, the suggested technique achieves an MSE of 0.005, which is much less than the 0.076 and 0.0025 MSEs obtained by EMD wavelet adaptive thresholding and soft thresholding correspondingly. This indicates that the proposed approach effectively eliminates noise while preserving significant signal characteristics, leading to an improved and less erroneous signal reconstruction. Furthermore, the proposed method outperformed conventional techniques and demonstrated improved noise reduction and signal clarity, achieving an SNR of 19.24 and a PRD of 20.38 at 10 dB SNR.

Enabling Seamless Interoperability of Digital Systems in Smart Cities Using API: A Systematic Literature Review

ABSTRACT Smart cities are envisioned to achieve an urban space where systems are connected to provide value added services to citizens and stakeholders, improve quality of life, and enhance institutional effectiveness. In smart cities, digital systems are deployed that comprise a wide range of service providers who collaborate to provide digital services to citizens. Hence, there is a need for accessible data that can be used to develop innovative applications and services to improve citizens’ quality of life. Unfortunately, achieving interoperability of digital systems is challenging due to isolated systems referred to as vertical silos, which cannot communicate with each other due to lack of standardization of interfaces. Thus, interoperability is seen as a barrier for the seamless transfer of data between systems deployed in smart cities. In order to facilitate interoperability of systems in smart cities, this article conducts a Systematic Literature Review (SLR) to investigate the role of Application Programming Interface (API) to enable seamless interoperability in smart cities. An SLR was employed to guide the selection of studies retrieved to adequately address the formulated research questions. Findings from this study present the importance of interoperability in smart urban environments and help in providing an overview and categorization of existing interoperability solutions to make cities smarter. The findings also identified prior interoperability approaches in smart cities. Findings present a case scenario that depicts how APIs support seamless interoperability among digital systems in a smart city environment. Lastly, the findings elaborate on available API management tools and further suggest that APIs can significantly contribute to improve interoperability between digital systems and decrease barriers in providing innovative digital services in smart cities.

Minimizing Delay and Power Consumption at the Edge.

Edge computing systems must offer low latency at low cost and low power consumption for sensors and other applications, including the IoT, smart vehicles, smart homes, and 6G. Thus, substantial research has been conducted to identify optimum task allocation schemes in this context using non-linear optimization, machine learning, and market-based algorithms. Prior work has mainly focused on two methodologies: (i) formulating non-linear optimizations that lead to NP-hard problems, which are processed via heuristics, and (ii) using AI-based formulations, such as reinforcement learning, that are then tested with simulations. These prior approaches have two shortcomings: (a) there is no guarantee that optimum solutions are achieved, and (b) they do not provide an explicit formula for the fraction of tasks that are allocated to the different servers to achieve a specified optimum. This paper offers a radically different and mathematically based principled method that explicitly computes the optimum fraction of jobs that should be allocated to the different servers to (1) minimize the average latency (delay) of the jobs that are allocated to the edge servers and (2) minimize the average energy consumption of these jobs at the set of edge servers. These results are obtained with a mathematical model of a multiple-server edge system that is managed by a task distribution platform, whose equations are derived and solved using methods from stochastic processes. This approach has low computational cost and provides simple linear complexity formulas to compute the fraction of tasks that should be assigned to the different servers to achieve minimum latency and minimum energy consumption.

Noise-Robust Biometric Authentication Using Infrared Periocular Images Captured from a Head-Mounted Display

This study proposes a biometric authentication method using infrared (IR)-based periocular images captured in virtual reality (VR) environments with head-mounted displays (HMDs). The widespread application of VR technology highlights the growing need for robust user authentication in immersive environments. To address this, the study introduces a novel periocular biometric authentication system optimized for HMD usage. Ensuring reliable authentication in VR environments necessitates overcoming significant challenges, including flicker noise and infrared reflection. Flicker noise, caused by alternating current (AC)-powered lighting, produces banding artifacts in images captured by rolling-shutter cameras, obstructing biometric feature extraction. Additionally, IR reflection generates strong light glare on the iris surface, degrading image quality and negatively impacting the model’s generalization performance and authentication accuracy. This study utilized the AffectiVR dataset, which includes noisy images, to address these challenges. In the preprocessing phase, iris reflections were removed, reducing the Equal Error Rate (EER) from 6.73% to 5.52%. Furthermore, incorporating a Squeeze-and-Excitation (SE) block to mitigate flicker noise and enhance model robustness resulted in a final EER of 6.39%. Although the SE block slightly increased the EER, it significantly improved the model’s ability to suppress noise and focus on critical periocular features, ensuring enhanced robustness in challenging VR environments. Heatmap analysis revealed that the proposed model effectively utilized periocular features, such as the skin around the eyes and eye contours, compared to prior approaches. This study establishes a crucial groundwork for advancing robust biometric authentication systems capable of overcoming noise challenges in next-generation immersive platforms.

A Modal Deconstruction of Löb Induction

We present a novel analysis of the fundamental Löb induction principle from guarded recursion. Taking advantage of recent work in modal type theory and univalent foundations, we derive Löb induction from a simpler and more conceptual set of primitives. We then capitalize on these insights to present Gatsby, the first guarded type theory capturing the rich modal structure of the topos of trees alongside Löb induction without immediately precluding canonicity or normalization. We show that Gatsby can recover many prior approaches to guarded recursion and use its additional power to improve on prior examples. We crucially rely on homotopical insights and Gatsby constitutes a new application of univalent foundations to the theory of programming languages.

Water vapor density field estimation using commercial microwave link attenuation combined with temperature measurements

Accurate water vapor density (WVD) measurement is critical for weather models, health risk management, and industrial management among many other applications. A number of machine-learning based algorithms (e.g. support vector machine) for estimating water vapor density at a reference weather station using the received signal level values measured at a commercial microwave link has been proposed in the past, and also was expanded to include a combination of three commercial microwave links with temperature measurements to achieve a higher estimation accuracy (with respect to the root mean square error at a given location). In this paper, we leverage on the preliminary potential presented, and propose enhanced machine learning models that utilize a larger number of CMLs combined with temperature data inside a given area to estimate a reference weather station humidity measurements. We then show how the presented approach can be expanded to estimate the water vapor density field - taking into consideration the elevation via the humidity-elevation profile. The models were evaluated using data from 32 weather stations and 505 CMLs in Germany, with performance assessed through root mean square error (RMSE) and correlation coefficients (CC). The enhanced models achieved a mean RMSE of 0.587 g/m³ for WVD field estimation, outperforming prior approaches as well as can be used as "virtual weather stations" - to estimate the water vapor density values in locations where no actual weather stations exist.

Enhancing energy efficiency and reliability in wireless sensor networks using BioGAT optimization

The BioGAT model, as proposed, presents a novel methodology for enhancing the efficiency of wireless sensor networks (WSNs), which are essential elements of contemporary communication and sensing systems. For real-time monitoring and data analysis, WSNs are comprised of autonomous sensor nodes that are outfitted with processing, wireless communication, and sensing capabilities. These nodes are deployed in a variety of environments. By means of an advanced optimization model, this work aims to address critical challenges in WSNs, specifically in the areas of node placement, energy efficiency, and network reliability. By utilizing biogeography-based optimization (BBO) and graph attention networks (GAT), the BioGAT model endeavors to dynamically adapt to network changes while achieving a balance between efficient coverage and energy consumption. Cluster heads (CHs), which are essential for the aggregation of data, have a significant impact on improvements in energy efficiency and the longevity of networks. By means of comprehensive simulations and evaluation, this study presents exceptional outcomes. The BioGAT model outperforms prior approaches by attaining a 95% packet delivery ratio and an enhanced throughput. In addition, the model effectively decreases mean energy consumption, underscoring its capacity to improve the sustainability and dependability of networks in a variety of WSN applications.

Phase‐Field Modeling of Fracture Under Compression and Confinement in Anisotropic Geomaterials

ABSTRACTStrongly anisotropic geomaterials, such as layered shales, have been observed to undergo fracture under compressive loading. This paper applies a phase‐field fracture model to study this fracture process. While phase‐field fracture models have several advantages—primarily that the fracture path is not predetermined but arises naturally from the evolution of a smooth non‐singular damage field—they provide unphysical predictions when the stress state is complex and includes compression that can cause crack faces to contact.Building on a recently developed phase‐field model that accounts for compressive traction across the crack face, this paper extends the model to the setting of anisotropic fracture. The key features of the model include the following: (1) a homogenized anisotropic elastic response and strongly anisotropic model for the work to fracture; (2) an effective damage response that accounts consistently for compressive traction across the crack face, that is derived from the anisotropic elastic response; (3) a regularized crack normal field that overcomes the shortcomings of the isotropic setting, and enables the correct crack response, both across and transverse to the crack face.To test the model, we first compare the predictions to phase‐field fracture evolution calculations in a fully resolved layered specimen with spatial inhomogeneity, and show that it captures the overall patterns of crack growth. We then apply the model to previously reported experimental observations of fracture evolution in laboratory specimens of shales under compression with confinement, and find that it predicts well the observed crack patterns in a broad range of loading conditions. We further apply the model to predict the growth of wing cracks under compression and confinement. Prior approaches to simulate wing cracks have treated the initial cracks as an external boundary, which makes them difficult to apply to general settings. Here, the effective crack response model enables us to treat the initial crack simply as a nonsingular damaged zone within the computational domain, thereby allowing for easy and general computations.

Data-Driven Temperature and Catalyst Optimization in Hydrogen Production using K-means Clustering

The optimization of hydrogen production as part of the transition to sustainable energy presents a challenge as the interplay of operational parameters decides the production rate such as temperature and the reaction performance of a catalyst. Typically, efficiency gains are sought through trial-and-error experimentation, or using linear models, which cannot model nonlinear relationships or high dimensional interactions. These issues must be dealt with advanced, data-driven techniques to uncover hidden patterns in large data sets. In this research K-means clustering, a machine learning technique, is used to analyze and optimize hydrogen production processes. The study used a dataset of 980 records derived from experimental results to discover five distinct catalyst behavior clusters based on different temperature conditions. The methodology consisted of data preprocessing, exploratory analysis, and model training and it was validated using the Elbow Method and silhouette scores. At the end, the optimal cluster configuration resulted in an SSD of 150.23 and a silhouette score of 0.72, suggesting good-quality clustering. We then used cluster analysis to derive operational patterns including temperature ranges for which certain catalysts operate at their optimum and unique performance profiles, to make more precise data-driven recommendations. Results show that machine learning is novel and superior to prior approaches to this problem, offering additional insight than traditional statistical methods. The approach allows for higher precision in the selection of the catalyst and in the process, optimization employing the identification of different operational profiles. Additional variables, like pressure and reaction times, could be added to this framework for future research, or adaptive clustering models could be used to improve real-time production optimization. This work demonstrates the utility of machine learning in the development of efficient and sustainable systems to produce hydrogen.

Enhanced performance of hafnia self-rectifying ferroelectric tunnel junctions at cryogenic temperatures

The advancement in high-performance computing technologies, including quantum and aerospace systems, necessitates components that operate efficiently at cryogenic temperatures. In this study, we demonstrate a hafnia-based ferroelectric tunnel junction (FTJ) that achieves a record-high tunneling electroresistance (TER) ratio of over 200,000 and decade-long retention characteristics. By introducing asymmetric oxygen vacancies through the strategic use of indium oxide (InOx) layer, we enhance the TER ratio without increasing off-current, addressing the longstanding issue of low on-current in hafnia-based FTJs. Unlike prior approaches that led to leakage currents, our method optimizes tunneling behavior by leveraging the differential oxygen dissociation energy between InOx and hafnium zirconium oxide (HZO). This results in asymmetric modulation of the tunnel barrier, enhancing electron tunneling in one polarization state while maintaining stability in the opposite state. Furthermore, we explore the intrinsic characteristics of the FTJ at cryogenic temperatures, where reduced thermal energy minimizes leakage currents and allows the maximization of device performance. These findings establish a new benchmark for TER in hafnia-based FTJs and provide valuable insights for the integration of these devices into advanced cryogenic memory systems.Graphical

Ensemble Deep Network for Secured Refactoring Framework by Predicting Code‐Bad Smells in Software Projects

ABSTRACTIn modern times, refactoring is one of the significantly utilized approaches for enhancing the software's quality like understandability, testability, and maintainability. Moreover, the refactoring effect on its security has been underrated. In addition to that, there are only a few studies that offer the classification over refactoring approaches depending on the effect over the quality attributes that help the designer to attain certain objectives by choosing the most significant approach and it is applied in the right places based on the specified software quality attributes. The contradictory outcomes are attained by considering the quality of the software creates limitations for the developers while performing the software refactoring process. In this paper, a secured deep learning‐based software refactoring approach is designed. At first, software projects collected from online sources are offered as input for this software refactoring process to detect the security metrics in the projects. After detecting the security metrics, refactoring is applied in the software projects to change the internal design. Then, the security metrics of the refactored projects are detected again. Further, the security metrics computed before and after refactoring are compared with the software projects. The projects are labeled based on security, needs, and refactoring level. Then, the Ensemble Attention‐based Deep Network (EA‐DNet) is developed, which is designed with the Recurrent Neural Network (RNN), Deep Temporal Convolution Network (DTCN), and Bi‐directional Long Short Term Memory (Bi‐LSTM). This network is trained to get better results in the prediction of code‐bad smells in software projects. The prior software refactoring approaches are compared with the proposed code‐bad smells‐based software refactoring process.

Initial State Preparation for Quantum Chemistry on Quantum Computers

Quantum algorithms for ground-state energy estimation of chemical systems require a high-quality initial state. However, initial state preparation is commonly either neglected entirely, or assumed to be solved by a simple product state like Hartree-Fock. Even if a nontrivial state is prepared, strong correlations render ground-state overlap inadequate for quality assessment. In this work, we address the initial state preparation problem with an end-to-end algorithm that and the quality of initial states, accomplishing the latter with a new metric—the energy distribution. To be able to prepare more complicated initial states, we introduce an implementation technique for states in the form of a sum of Slater determinants that exhibits significantly better scaling than all prior approaches. We also propose low-precision quantum phase estimation (QPE) for further state quality refinement. The complete algorithm is capable of generating high-quality states for energy estimation, and is shown in select cases to lower the overall estimation cost by several orders of magnitude when compared with the best single product state ansatz. More broadly, the energy distribution picture suggests that the goal of QPE should be reinterpreted as generating improvements compared to the energy of the initial state and other classical estimates: such an improvement can still be achieved even if QPE does not project directly onto the ground state. Finally, we show how the energy distribution can help in identifying potential quantum advantage. Published by the American Physical Society 2024

Mapping Anatomical Boundaries from Histology to Antemortem MRI to Inform Automatic Segmentation of Medial Temporal Lobe Subregions

AbstractBackgroundThe medial temporal lobe (MTL) is the epicenter of both primary and concomitant molecular pathologies in Alzheimer’s disease (AD). The intricate anatomy of the MTL has been the subject of extensive study over the past two centuries. However, current PET and MRI AD biomarkers use often crude parcellations of the MTL that have not been sufficiently validated vis‐à‐vis anatomical ground truth. Here we use a unique dataset of finely annotated serial histology, ex vivo MRI and antemortem 3T in vivo MRI of the MTL from 17 brain donors to train and evaluate an automatic MTL subregion segmentation algorithm. To our knowledge, this is the most comprehensive attempt to infuse MRI biomarkers with cytoarchitectural reference annotations.MethodA team of neuroanatomists annotated the boundaries of 27 MTL subregions on over 2000 serial Nissl histological sections from 17 brain donors. These boundaries were mapped to 9.4T ex vivo MRI (proton density, 0.2x0.2x0.2mm3) and, subsequently, to 3T in vivo MRI (T2‐weighted, ∼0.4x0.4x2.6mm3) using deformable registration, with extensive manual editing after each registration step to correct for registration errors and ensure 3D continuity and smoothness (Figure 1). Deep learning method nnU‐Net was trained on the resulting in vivo MRI annotations to automatically segment a set of major subregions (formed by merging smaller subregions). Surface‐based registration between an MTL template and the output of nnU‐Net was used to parcellate major subregions into 27 smaller subregions. Segmentation accuracy was assessed by five‐fold cross‐validation.ResultFigure 2 compares nnU‐Net and surface‐based automatic segmentations with corresponding cytoarchitectural reference segmentations. Table 1 reports segmentation accuracy for major subregions in terms of Dice coefficient. While accuracy is not as high as in some previous MTL subregion segmentation approaches, this is not surprising because the anatomical variability of cytoarchitecture‐based ground truth annotations in our approach is likely much higher than in prior approaches where ground truth was generated by applying heuristic/geometric rules.ConclusionIt is feasible to leverage cytoarchitecturally defined anatomical boundaries for automatic in vivo MRI segmentation. However, high variability in the location of cytoarchitectural borders poses clear limitations on MTL segmentation accuracy.

The Relationship of Neuropsychiatric Symptoms and ADRD Pathologies

AbstractBackgroundNeuropsychiatric symptoms (NPS) in the Alzheimer’s Disease (AD) clinical spectrum are common, yet the pathological underpinnings are unclear. Prior research has been inconsistent, attributed in part to concomitant co‐pathologies. Additionally, prior modeling approaches have disregarded the distributional properties of right‐skewed NPS data and utilized predominantly racially homogenous cohorts. Thus, we examined the relationship between NPS and markers of vascular injury and neurodegeneration using generalized linear models (GLM) in a diverse community‐based sample. Of additional interest was whether these relationships differed by cognitive status or race.MethodsWe evaluated cross‐sectional relationships of brain white matter hyperintensity volume (WMHv) and plasma neurofilament light chain (pNfL) to NPS (NPI‐Q total severity scores) in 588 Wake Forest ADRC Clinical Core study participants. Analyses included participants with consensus diagnoses of cognitively unimpaired (CU; N = 348) and mild cognitive impairment (MCI; N = 240). WMHv and pNfL were log‐transformed prior to model entry. GLMs included age, sex, education, race, and cognitive status as covariates and pathologies as predictor variables. Secondary models included aforementioned covariates and pathology by (a) cognitive status and (b) race interaction terms to investigate how these associations differed.ResultsWMHv was not significantly associated with NPS (N = 561; β = 0.203; p = 0.053); further, we did not observe an interaction between cognitive status and WMHv (N = 561; β = ‐0.055; p = 0.800). However, a significant race by WMHv interaction on NPI‐Q was demonstrated (N = 561; β = ‐0.414; p = 0.016). In White participants, elevated WMHv was significantly related to elevated NPS (N = 434; β = 0.325; p = 0.006); whereas in Black/African American participants elevated WMHv was significantly related to lower NPS (N = 119; β = ‐0.239; p = 0.049). pNfL was not significantly associated with NPS (N = 506; β = 0.149; p = 0.576). A significant pNfL by cognitive status interaction on NPI‐Q was found (N = 506; β = ‐1.055; p = 0.029; Figure 2), where elevated pNfL was related to elevated NPS in MCI participants (N = 198; β = 1.656; p = 0.004) but not CU participants (N = 308; β = ‐0.316; p = 0.260). However, we did not observe an interaction between race and pNfL (N = 506; β = 0.185; p = 0.678).ConclusionConsistent with previous work we demonstrate associations of pNfL and WMHv with NPS. Future analyses with this cohort will examine additional biomarkers, including amyloid‐beta and tau measured in CSF, plasma, and PET.

The Relationship of Neuropsychiatric Symptoms and ADRD Pathologies

AbstractBackgroundNeuropsychiatric symptoms (NPS) in the Alzheimer’s Disease (AD) clinical spectrum are common, yet the pathological underpinnings are unclear. Prior research has been inconsistent, attributed in part to concomitant co‐pathologies. Additionally, prior modeling approaches have disregarded the distributional properties of right‐skewed NPS data and utilized predominantly racially homogenous cohorts. Thus, we examined the relationship between NPS and markers of vascular injury and neurodegeneration using generalized linear models (GLM) in a diverse community‐based sample. Of additional interest was whether these relationships differed by cognitive status or race.MethodsWe evaluated cross‐sectional relationships of brain white matter hyperintensity volume (WMHv) and plasma neurofilament light chain (pNfL) to NPS (NPI‐Q total severity scores) in 588 Wake Forest ADRC Clinical Core study participants. Analyses included participants with consensus diagnoses of cognitively unimpaired (CU; N = 348) and mild cognitive impairment (MCI; N = 240). WMHv and pNfL were log‐transformed prior to model entry. GLMs included age, sex, education, race, and cognitive status as covariates and pathologies as predictor variables. Secondary models included aforementioned covariates and pathology by (a) cognitive status and (b) race interaction terms to investigate how these associations differed.ResultsWMHv was not significantly associated with NPS (N = 561; ß = 0.203; p = 0.053); further, we did not observe an interaction between cognitive status and WMHv (N = 561; ß = ‐0.055; p = 0.800). However, a significant race by WMHv interaction on NPI‐Q was demonstrated (N = 561; ß = ‐0.414; p = 0.016). In White participants, elevated WMHv was significantly related to elevated NPS (N = 434; ß = 0.325; p = 0.006); whereas in Black/African American participants elevated WMHv was significantly related to lower NPS (N = 119; ß = ‐0.239; p = 0.049). pNfL was not significantly associated with NPS (N = 506; ß = 0.149; p = 0.576). A significant pNfL by cognitive status interaction on NPI‐Q was found (N = 506; ß = ‐1.055; p = 0.029; Figure 2), where elevated pNfL was related to elevated NPS in MCI participants (N = 198; ß = 1.656; p = 0.004) but not CU participants (N = 308; ß = ‐0.316; p = 0.260). However, we did not observe an interaction between race and pNfL (N = 506; ß = 0.185; p = 0.678).ConclusionConsistent with previous work we demonstrate associations of pNfL and WMHv with NPS. Future analyses with this cohort will examine additional biomarkers, including amyloid‐beta and tau measured in CSF, plasma, and PET.

Quantum differential cryptanalysis based on Bernstein-Vazirani algorithm

Recent research has demonstrated the potential of quantum algorithms to exploit vulnerabilities in various popular constructions, such as certain block ciphers like Feistel, Even-Mansour, and multiple MACs, within the superposition query model. In this study, we delve into the security of block ciphers against quantum threats, particularly investigating their susceptibility to cryptanalysis techniques, notably exploring quantum adaptations of differential cryptanalysis. Initially, we introduce a BV-based quantum algorithm for identifying linear structures with a complexity of O(n), where n denotes the number of bits in the function. Subsequently, we illustrate the application of this algorithm in devising quantum differential cryptanalysis techniques, including quantum differential cryptanalysis, quantum small probability differential cryptanalysis, and quantum impossible differential cryptanalysis, demonstrating polynomial acceleration compared to prior approaches. By treating the encryption function as a unified entity, our algorithm circumvents the traditional challenge of extending differential paths in differential cryptanalysis.

Security Mechanism in MAMATA Healthcare System Using Rule based Algorithm for Maternal Hospitals and Pathology Laboratories

Recently, innovation in the digital health industry has grown steadily, trending with medical cyber-physical systems’ (MCPS) integrated solutions, regulations, real-time analytics, and software frameworks. Amid these advancements, the paramount concern is the security of medical information, particularly in maternity hospitals where data breaches must be rigorously avoided. While some researchers have explored security methods, they have encountered limitations and identified existing risks. In this research article, the authors propose a novel security mechanism/method called MAMATA Agent/BOT (i.e. Medical-data-Access-for-Maternal-Treat-ment-in-rural-Area) to discover and mitigate trending vulnerabilities, drawing reference from the OWASP (Open Web Application Security Project) Top 10 Report 2017/2020. The author’s approach leverages a rule-based algorithm, enhancing the robustness of the security mechanism. This MCPS system and security mechanism are implemented within PHP and Laravel environments, and comparative analysis is carried out against a few existing security methods commonly employed in general healthcare (like SOA/SOD Framework, Andrew Austin test, Marcelo test, etc..). Results demonstrate a quantified improvement, e.g., 60-70% improvement in detectable vulnerabilities compared to prior approaches. Additionally, the authors extend their previous research by depicting different turns of events and applying a more stable, secured structure tailored to medical applications. Quantitative data presented with comparative analysis in tabular format validates the effectiveness of the proposed methods by detecting and, or fixing 7 to 9 major vulnerabilities (out of the top 10) identified by security assessment tools like ‘Burp Suite Proxy’ and ‘Nikto’.

Federated Learning Client Pruning for Noisy Labels

Federated Learning (FL) enables collaborative model training across decentralized edge devices while preserving data privacy. However, existing FL methods often assume clean annotated datasets, impractical for resource-constrained edge devices. In reality, noisy labels are prevalent, posing significant challenges to FL performance. Prior approaches attempt label correction and robust training techniques but exhibit limited efficacy, particularly under high noise levels. This paper introduces ClipFL ( F ederated L earning Cli ent P runing), a novel framework addressing noisy labels from a fresh perspective. ClipFL identifies and excludes noisy clients based on their performance on a clean validation dataset, tracked using a Noise Candidacy Score (NCS). The framework comprises three phases: pre-client pruning to identify potential noisy clients and calculate their NCS, client pruning to exclude a percentage of clients with the highest NCS, and post-client pruning for fine-tuning the global model with standard FL on clean clients. Empirical evaluation demonstrates ClipFL’s efficacy across diverse datasets and noise levels, achieving accurate noisy client identification, superior performance, faster convergence, and reduced communication costs compared to state-of-the-art FL methods. Our code is available at https://github.com/MMorafah/ClipFL.

Using Short Molecular Dynamics Simulations to Determine the Important Features of Interactions in Antibody-Protein Complexes.

The last few years have seen the rapid proliferation of machine learning methods to design binding proteins. Although these methods have shown large increases in experimental success rates compared to prior approaches, the majority of their predictions fail when they are experimentally tested. It is evident that computational methods still struggle to distinguish the features of real protein binding interfaces from false predictions. Short molecular dynamics simulations of 20 antibody-protein complexes were conducted to identify features of interactions that should occur in binding interfaces. Intermolecular salt bridges, hydrogen bonds, and hydrophobic interactions were evaluated for their persistences, energies, and stabilities during the simulations. It was found that only the hydrogen bonds where both residues are stabilized in the bound complex are expected to persist and meaningfully contribute to binding between the proteins. In contrast, stabilization was not a requirement for salt bridges and hydrophobic interactions to persist. Still, interactions where both residues are stabilized in the bound complex persist significantly longer and have significantly stronger energies than other interactions. Two hundred and twenty real antibody-protein complexes and 8194 decoy complexes were used to train and test a random forest classifier using the features of expected persistent interactions identified in this study and the macromolecular features of interaction energy (IE), buried surface area (BSA), IE/BSA, and shape complementarity. It was compared to a classifier trained only on the expected persistent interaction features and another trained only on the macromolecular features. Inclusion of the expected persistent interaction features reduced the false positive rate of the classifier by two- to five-fold across a range of true positive classification rates.

NetGSR: Towards Efficient and Reliable Network Monitoring with Generative Super Resolution

Network monitoring systems are a key building block in today's networks. They all follow a common framework where measurement data from network elements is aggregated at a central collector for network-wide visibility. When designing network monitoring systems, two key properties have to be taken into account: (1) efficiency, to minimize the communication overhead from network elements to the collector; (2) high-fidelity, to faithfully represent the network status. However, in presence of network dynamics, tracking the right operating point to ensure both high fidelity and efficiency is hard and we observe that prior monitoring approaches trade off one for the other. In this paper, we show that it is possible to satisfy both these properties with NetGSR, a new deep learning based solution we introduce that reconstructs the fine-grained behavior of network status at the collector while requiring low resolution measurement data from network elements. This is achieved through a combination of a new custom-tailored conditional deep generative model (DistilGAN), and a new feedback mechanism (Xaminer) based on model uncertainty estimation and denoising that allows the collector to adjust the sampling rate for measurement data from network elements, at run-time. We extensively evaluate NetGSR using three different network scenarios with corresponding real-world network monitoring datasets as well as two downstream use cases. We show that NetGSR can faithfully reconstruct fine-grained network status with 25x greater measurement efficiency than prior approaches while requiring only few ms of inference time at the collector.