Local Machine Research Articles

PILE: Robust Privacy-Preserving Federated Learning Via Verifiable Perturbations

Federated learning (FL) protects training data in clients by collaboratively training local machine learning models of clients for a global model, instead of directly feeding the training data to the server. However, existing studies show that FL is vulnerable to various attacks, resulting in training data leakage or interfering with the model training. Specifically, an adversary can analyze local gradients and the global model to infer clients' data, and poison local gradients to generate an inaccurate global model. It is extremely challenging to guarantee strong privacy protection of training data while ensuring the robustness of model training. None of the existing studies can achieve the goal. In this paper, we propose a robust privacy-preserving federated learning framework (PILE), which protects the privacy of local gradients and global models, while ensuring their correctness by gradient verification where the server verifies the computation process of local gradients. In PILE, we develop a verifiable perturbation scheme that makes confidential local gradients verifiable for gradient verification. In particular, we build two building blocks of zero-knowledge proofs for the gradient verification without revealing both local gradients and global models. We perform rigorous theoretical analysis that proves the security of PILE and evaluate PILE on both passive and active membership inference attacks. The experiment results show that the attack accuracy under PILE is between <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$[50.3\%,50.9\%]$</tex-math></inline-formula> , which is close to the random guesses. Particularly, compared to prior defenses that incur the accuracy losses ranging from <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$2\%$</tex-math></inline-formula> to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$13\%$</tex-math></inline-formula> , the accuracy loss of PILE is negligible, i.e., only <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\pm 0.3\%$</tex-math></inline-formula> accuracy loss.

The association between BMI and serum uric acid is partially mediated by gut microbiota.

Obesity is a risk factor for the development of hyperuricemia, both of which were related to gut microbiota. However, whether alterations in the gut microbiota lie in the pathways mediating obesity's effects on hyperuricemia is less clear. Body mass index (BMI) and serum uric acid (SUA) were separately important indicators of obesity and hyperuricemia. Our study aims to investigate whether BMI-related gut microbiota characteristics would mediate the association between BMI and SUA levels. A total of 6,280 participants from Guangdong Gut Microbiome Project were included in this study. Stool samples were collected for 16S rRNA gene sequencing. The results revealed that BMI was significantly and positively associated with SUA. Meanwhile, BMI was significantly associated with the abundance of 102 gut microbial genera, 16 of which were also significantly associated with SUA. The mediation analysis revealed that the association between BMI and SUA was partially mediated by the abundance of Proteobacteria (proportion mediated: 0.94%, P < 0.05). At the genus level, 25 bacterial genera, including Ralstonia, Oscillospira, Faecalibacterium, etc., could also partially mediate the association of BMI with SUA (the highest proportion is mediated by Ralstonia, proportion mediated: 2.76%, P < 0.05). This study provided evidence for the associations among BMI, gut microbiota, and SUA, and the mediation analysis suggested that the association of BMI with SUA was partially mediated by the gut microbiota. IMPORTANCE Using 16S rRNA sequencing analysis, local interpretable machine learning technique analysis and mediation analysis were used to explore the association between BMI with SUA, and the mediating effects of gut microbial dysbiosis in the association were investigated.

Distributed debiased estimation of high-dimensional partially linear models with jumps

In this paper, we focus on the estimations of both parameter vector and nonparametric component in a high-dimensional partially linear model with jumps within the framework of divide and conquer strategy. We find that a three-stage estimation procedure works well in this setting. Applying the lasso penalty and projected spline approximation, first a profiled estimator for the linear part and a projected spline estimator for the nonparametric part are obtained on each local machine. In the second stage, an efficient jump detection algorithm is developed to obtain the new knot sequence, and then based on this, the estimate of the nonparametric function is obtained and averaged after plugging in the linear part estimate on each local machine. The aggregated estimate of the nonparametric function is then computed by pooling these local estimates. In the third stage, a debiased lasso estimator is averaged to obtain a distributed debiased estimator of the linear part after plugging in the aggregated estimate of nonparametric function. Asymptotic properties of resultant estimators are established under some mild assumptions. Some simulations are conducted to illustrate the empirical performances of our proposed method.

Edge assignment in edge federated learning

Federated Learning (FL) is a recent Machine Learning method for training with private data locally stored in distributed machines without gathering them into one place for central learning. Because FL depends on a central server for repeated aggregation of local training models, this server is prone to become a performance bottleneck. Therefore, one can combine FL with Edge Computing: introduce a layer of edge servers to each serve as a regional aggregator to offload the main server. The scalability is thus improved, however at the cost of learning accuracy. We show that this cost can be alleviated with a proper choice of edge server assignment: which edge servers should aggregate the training models from which local machines. In this paper, we propose an assignment solution for this purpose. Our solution is especially useful for the case of non-IID training data which is well-known to hinder today’s FL performance. Our findings are substantiated with an evaluation study using real-world datasets.

Hybrid feature extraction and LLTSA-based dimension reduction for vein pattern recognition

ABSTRACT In information and security, the personal identification of individuals becomes much more important. For improving security, several biometric recognition techniques are implemented. However, in finger vein recognition, it faces the critical problem of fake finger vein images, security and less accuracy. To conquer this problem, Hybrid Feature Extraction with Linear Local Tangent Space Alignment-based dimension reduction and Support Vector Machine classifier (HFE–LLTSA–SVM) is proposed. In this hybrid, FE is considered as the combination of histogram of oriented gradients (HOG), grey-level co-occurrence matrix (GLCM), stationary wavelet transform (SWT), and local binary pattern (LBP) for extracting the hybrid feature. LLTSA perform dimension reduction in the outputs of HFE from HOG, GLCM, and LBP. Furthermore, SVM is used for classification which gives authentication based on error-correcting code. Finally, the performance parameters were calculated and the proposed method achieved better accuracy of 99.75%, when compared with existing methods.

Automated Age-Related Macular Degeneration Detector on Optical Coherence Tomography Images Using Slice-Sum Local Binary Patterns and Support Vector Machine.

Artificial intelligence has revolutionised smart medicine, resulting in enhanced medical care. This study presents an automated detector chip for age-related macular degeneration (AMD) using a support vector machine (SVM) and three-dimensional (3D) optical coherence tomography (OCT) volume. The aim is to assist ophthalmologists by reducing the time-consuming AMD medical examination. Using the property of 3D OCT volume, a modified feature vector connected method called slice-sum is proposed, reducing computational complexity while maintaining high detection accuracy. Compared to previous methods, this method significantly reduces computational complexity by at least a hundredfold. Image adjustment and noise removal steps are excluded for classification accuracy, and the feature extraction algorithm of local binary patterns is determined based on hardware consumption considerations. Through optimisation of the feature vector connection method after feature extraction, the computational complexity of SVM detection is significantly reduced, making it applicable to similar 3D datasets. Additionally, the design supports model replacement, allowing users to train and update classification models as needed. Using TSMC 40 nm CMOS technology, the proposed detector achieves a core area of 0.12 mm2 while demonstrating a classification throughput of 8.87 decisions/s at a maximum operating frequency of 454.54 MHz. The detector achieves a final testing classification accuracy of 92.31%.

Oracle text positioning system based on improved YOLOv7-FC model

In recent years, how to use computer to efficiently and accurately recognize oracle characters on oracle rubbings has become the focus of research. Due to the severe damage caused by the natural environment and the characteristics of complex background and loud noise, the oracle bone rubbings are difficult to separate the foreground and back of the oracle bone inscriptions. Therefore, it will be the future research direction to eliminate the negative impact of complex background and accurately locate the Oracle bone inscriptions. This paper puts forward the YOLOv7-FC neural network model and realizes the specific application in the oracle topology data set, and presents the system of single oracle character positioning results.The experimental results show that on the HWOBC-A dataset, the mAP value of the YOLOv7-FC neural network model proposed in this paper reaches 0.937. Compared with the original YOLOv7 model, it has improved by 0.22, effectively balancing the relationship between running speed and average accuracy, and providing feasibility for Oracle text localization machine learning method.

Spatial attention-based residual network for human burn identification and classification

Diagnosing burns in humans has become critical, as early identification can save lives. The manual process of burn diagnosis is time-consuming and complex, even for experienced doctors. Machine learning (ML) and deep convolutional neural network (CNN) models have emerged as the standard for medical image diagnosis. The ML-based approach typically requires handcrafted features for training, which may result in suboptimal performance. Conversely, DL-based methods automatically extract features, but designing a robust model is challenging. Additionally, shallow DL methods lack long-range feature dependency, decreasing efficiency in various applications. We implemented several deep CNN models, ResNeXt, VGG16, and AlexNet, for human burn diagnosis. The results obtained from these models were found to be less reliable since shallow deep CNN models need improved attention modules to preserve the feature dependencies. Therefore, in the proposed study, the feature map is divided into several categories, and the channel dependencies between any two channel mappings within a given class are highlighted. A spatial attention map is built by considering the links between features and their locations. Our attention-based model BuRnGANeXt50 kernel and convolutional layers are also optimized for human burn diagnosis. The earlier study classified the burn based on depth of graft and non-graft. We first classified the burn based on the degree. Subsequently, it is classified into graft and non-graft. Furthermore, the proposed model performance is evaluated on Burns_BIP_US_database. The sensitivity of the BuRnGANeXt50 is 97.22% and 99.14%, respectively, for classifying burns based on degree and depth. This model may be used for quick screening of burn patients and can be executed in the cloud or on a local machine. The code of the proposed method can be accessed at https://github.com/dhirujis02/Journal.git for the sake of reproducibility.

INTERNATIONAL PRACTICE ASSISTANCE AT REGIONAL REVENUE DEPARTMENT OF TTU DISTRICT BY STUDENTS OF THE UNIVERSITY OF TIMOR UNIVERSITY DEVELOPMENT AND ECONOMIC MANAGEMENT STUDY PROGRAM

This community service is carried out by assisting apprenticeship practices at the Regional Revenue Service Office of North Central Timor Regency. With the aim of providing opportunities for students to apply theory and knowledge during lectures to practical realities in the world of work. The method used is the Field Work Practice at the TTU Regional Revenue Service Office. Assistance is carried out by 4 students with an effective period of 50 working days. As for the results of this internship work practice, apprentices can learn directly about the use of local tax applications, attend PAD evaluation meetings, assist the head of the billboard-themed data collection section, file taxes, filing incoming and outgoing letters, operate a local tax receipt printing machine, recapitulate mineral tax determination. and non-metal as well as assistance from TTU Regency.

Semi-profiled distributed estimation for high-dimensional partially linear model

As a leading example in the semiparametric model, the partially linear model is prominent in modeling complex data. The challenges associated with designing an efficient estimation algorithm in the distributed environment are not yet well addressed. The existing works require a constraint on the number of local machines to guarantee the optimality of the global estimator. In addition, a multi-round profiled estimator will lead to huge communication complexity. To further reduce communication costs, a novel semi-profiled estimation procedure is proposed, which provides an iterative skeleton to reduce estimation error. A new multi-round distributed algorithm based on the centralized semi-profiled estimator is developed, which can estimate the parametric and nonparametric simultaneously. The theoretical results indicate that the corresponding convergence rates can achieve the optimal orders within a constant number of communication rounds. The advantages of the proposed estimation method are demonstrated via numerical experiments on synthetic and real data.

Simulation research on high speed centrifugal residual stress homogenization of turbine disk

At present, problems such as excessive residual stress, local stress concentration and machining deformation are common encountered in the production and processing of turbine disk. In order to further improve the machining quality and performance of turbine disk components, a high speed centrifugal stress homogenization and strengthening method with in-depth research and application value is proposed in this paper. The simulation research on the centrifugal stress homogenization effect of turbine disk forging blank is carried out, and the distribution of residual stress at different speeds is obtained, the influence of rotating speed on stress homogenization effect is summarized.

Unipept Desktop 2.0: Construction of Targeted Reference Protein Databases for Metaproteogenomics Analyses.

Unipept Desktop 2.0 is the most recent iteration of the Unipept Desktop tool that adds support for the analysis of metaproteogenomics datasets. Unipept Desktop now supports the automatic construction of targeted protein reference databases that only contain proteins (originating from the UniProtKB resource) associated with a predetermined list of taxa. This improves both the taxonomic and functional resolution of a metaproteomic analysis and yields several technical advantages. By limiting the proteins present in a reference database, it is also possible to perform (meta)proteogenomics analyses. Since the protein reference database resides on the user's local machine, they have complete control over the database used during an analysis. Data no longer need to be transmitted over the Internet, decreasing the time required for an analysis and better safeguarding privacy-sensitive data. As a proof of concept, we present a case study in which a human gut metaproteome dataset is analyzed with Unipept Desktop 2.0 using different targeted databases based on matched 16S rRNA gene sequencing data.

Local Justice and Machine Learning: Modeling and Inferring Dynamic Ethical Preferences toward Allocations

We consider a setting in which a social planner has to make a sequence of decisions to allocate scarce resources in a high-stakes domain. Our goal is to understand stakeholders' dynamic moral preferences toward such allocational policies. In particular, we evaluate the sensitivity of moral preferences to the history of allocations and their perceived future impact on various socially salient groups. We propose a mathematical model to capture and infer such dynamic moral preferences. We illustrate our model through small-scale human-subject experiments focused on the allocation of scarce medical resource distributions during a hypothetical viral epidemic. We observe that participants' preferences are indeed history- and impact-dependent. Additionally, our preliminary experimental results reveal intriguing patterns specific to medical resources---a topic that is particularly salient against the backdrop of the global covid-19 pandemic.

A method for computing inverse parametric PDE problems with random-weight neural networks

We present a method for computing the inverse parameters and the solution field to inverse parametric partial differential equations (PDE) based on randomized neural networks. This extends the local extreme learning machine technique originally developed for forward PDEs to inverse problems. We develop three algorithms for training the neural network to solve the inverse PDE problem. The first algorithm (termed NLLSQ) determines the inverse parameters and the trainable network parameters all together by the nonlinear least squares method with perturbations (NLLSQ-perturb). The second algorithm (termed VarPro-F1) eliminates the inverse parameters from the overall problem by variable projection to attain a reduced problem about the trainable network parameters only. It solves the reduced problem first by the NLLSQ-perturb algorithm for the trainable network parameters, and then computes the inverse parameters by the linear least squares method. The third algorithm (termed VarPro-F2) eliminates the trainable network parameters from the overall problem by variable projection to attain a reduced problem about the inverse parameters only. It solves the reduced problem for the inverse parameters first, and then computes the trainable network parameters afterwards. VarPro-F1 and VarPro-F2 are reciprocal to each other in some sense. The presented method produces accurate results for inverse PDE problems, as shown by the numerical examples herein. For noise-free data, the errors of the inverse parameters and the solution field decrease exponentially as the number of collocation points or the number of trainable network parameters increases, and can reach a level close to the machine accuracy. For noisy data, the accuracy degrades compared with the case of noise-free data, but the method remains quite accurate. The presented method has been compared with the physics-informed neural network method.

Localized sinking electrochemical machining of pre-shaped convex structure on revolution surface using specifically designed cathode tool

Counter-rotating electrochemical machining (CRECM) is an effective method for machining thin-walled revolving parts with complex structures. However, it is difficult to achieve a desirable uniform straight sidewall during the CRECM of isolated convex structures with variable cross-sectional profiles, so a subsequent local machining process is necessary. In this paper, a localized sinking ECM method is proposed to locally machine a pre-shaped convex structure by CRECM. A mathematical model is established to analyze the anode shaping process. In addition, a special cathode tool is designed to control the junction mark between the subsequent dissolution regions and the surrounded non-machined areas. The simulation and experimental results indicate that the sidewall profiles of the convex structure can be improved remarkably. In lateral view, convex side wall taper is reduced from 21.81° to 3.05°, In front view, convex side wall taper is reduced from 6.45° to 0.28°, Furthermore, the junction mark can be effectively controlled to within 0.07 mm.

Information-theoretic privacy in federated submodel learning

We consider information-theoretic privacy in federated submodel learning, where a global server has multiple submodels. Compared to the privacy considered in the conventional federated submodel learning where secure aggregation is adopted for ensuring privacy, information-theoretic privacy provides the stronger protection on submodel selection by the local machine. We propose an achievable scheme that partially adopts the conventional private information retrieval (PIR) scheme that achieves the minimum amount of download. With respect to computation and communication overhead, we compare the achievable scheme with a naïve approach for federated submodel learning with information-theoretic privacy.

Mech2d: An Efficient Tool for High-Throughput Calculation of Mechanical Properties for Two-Dimensional Materials.

Two-dimensional (2D) materials have been a research hot topic in the passed decades due to their unique and fascinating properties. Among them, mechanical properties play an important role in their application. However, there lacks an effective tool for high-throughput calculating, analyzing and visualizing the mechanical properties of 2D materials. In this work, we present the mech2d package, a highly automated toolkit for calculating and analyzing the second-order elastic constants (SOECs) tensor and relevant properties of 2D materials by considering their symmetry. In the mech2d, the SOECs can be fitted by both the strain-energy and stress-strain approaches, where the energy or strain can be calculated by a first-principles engine, such as VASP. As a key feature, the mech2d package can automatically submit and collect the tasks from a local or remote machine with robust fault-tolerant ability, making it suitable for high-throughput calculation. The present code has been validated by several common 2D materials, including graphene, black phosphorene, GeSe2 and so on.

Target Tracking Algorithm Based on Generalized Regression Neural Network for Passive Bistatic Radar

In passive bistatic radar, high-performance target tracking is of crucial importance to a variety of applications. However, most existing tracking algorithms cannot achieve satisfactory performance due to poor azimuth estimation and the nonlinear relationship between target position and measurements. This paper proposes a novel tracking algorithm based on improved generalized regression neural networks (GRNN) and Kalman filtering (KF). First, the dependency between measurements and the target position is modeled by GRNN regression. A cooperative co-evolutionary genetic algorithm (CCGA) is then introduced to optimize the sample size and smoothing factor of the GRNN model to improve the performance. Finally, this paper applies KF to CCGA-GRNN localization results and forms the smoothing trajectories of the targets. Simulation and experimental results show that the performance of the proposed algorithm is better than that of the geometric localization (GL), GRNN, and extreme learning machine (ELM) based tracking approaches. Furthermore, field experiment data verifies the feasibility of the proposed algorithm.

Study of Global Photospheric and Chromospheric Flows Using Local Correlation Tracking and Machine Learning Methods I: Methodology and Uncertainty Estimates

Cyclical variations of the solar magnetic fields, and hence the level of solar activity, are among the top interests of space weather research. Surface flows in global-scale, in particular differential rotation and meridional flows, play important roles in the solar dynamo that describes the origin and variation of solar magnetic fields. In principle, differential rotation is the fundamental cause of dipole field formation and emergence, and meridional flows are the surface component of a longitudinal circulation that brings decayed field from low latitudes to polar regions. Such flows are key inputs and constraints of observational and modeling studies of solar cycles. Here, we present two methods, local correlation tracking (LCT) and machine learning-based self-supervised optical flow methods, to measure differential rotation and meridional flows from full-disk magnetograms that probe the photosphere and text{H}alpha images that probe the chromosphere, respectively. LCT is robust in deriving photospheric flows using magnetograms. However, we found that it failed to trace flows using time-sequence text{H}alpha data because of the strong dynamics of traceable features. The optical flow methods handle text{H}alpha data better to measure the chromospheric flow fields. We found that the differential rotation from photospheric and chromospheric measurements shows a strong correlation with a maximum of 2.85~upmu text{rad},text{s}^{-1} at the equator and the accuracy holds until 60^{circ } for the MDI and text{H}alpha, 75^{circ } for the HMI dataset. On the other hand, the meridional flow deduced from the chromospheric measurement shows a similar trend as the concurrent photospheric measurement within 60^{circ } with a maximum of 20~text{m},text{s}^{-1} at 40^{circ } in latitude. Furthermore, the measurement uncertainties are discussed.

Spatial prediction of soil salinity based on the Google Earth Engine platform with multitemporal synthetic remote sensing images

Accurately analyzing the spatial distribution of salinity at the regional scale is crucial for regional sustainable development. Currently, there are no reliable approaches for mapping salinity due to the substantial spatial variability of soil salinity. However, remote sensing technology has been successfully used to monitor soil properties in various locations. This research investigates an approach to choose the most appropriate remote sensing time window using a cloud computing platform and synthesizes multitemporal remote sensing images for high spatial resolution digital mapping of salinization at the regional scale. The research site for this experiment is the Werigan-Kuqa Oasis in Xinjiang, China. In the first stage, we screen the Landsat-8 surface reflectance datasets for four-time windows: April–October, May–September, June–August multitemporal, and July single-date images using the Google Earth Engine (GEE) cloud platform. Then, we synthesize images based on minimum, maximum, mean, and median values and construct multiple spectral indices. We completed model training and digital mapping in the second stage using a geemap. For the first time, we used geemap to integrate a local scikit-learn machine learning library with GEE to train a machine learning model. Next, we performed salinity prediction and mapping based on the best data set. This study demonstrates that (1) Various environmental variables contribute differently to modeling in the same time window. Additionally, these variables exhibit variation across different time windows. However, certain indices, including clay index (CLEX), near infrared (NIR), difference vegetation index (DVI), carbonate index (CAEX), green ratio vegetation index (GRVI), canopy response salinity (CRSI), and elevation, demonstrate stable contributions across different time windows. (2) Employing GEE to choose the suitable time window for synthesizing remote sensing images has a better prediction effect compared to single-date image prediction accuracy. Through experimental comparison, the best time window is May–September. (3) The mean synthesis approach from May to September exhibits the best performance in the model, while the median synthesis approach exhibits stable performance in the other two individual time windows. (4) Combining scikit-learn with GEE using Geemap improves model optimization and enhances the mapping performance. This study broadens the application of GEE in soil mapping and improves salinity prediction using multitemporal synthetic pictures with time frames.