SmartFed: Federated learning with collaborative optimization of the model algorithm and data enhancement

Near infrared (NIR) sensing technology has been widely implemented in various areas due to its noninvasive, green, and rapid measurement features. Artificial Neural Network (ANN) was used for the NIR calibration model. ANN can provide a reliable local calibration model using local data. However, a reliable local calibration model could be invalid or degraded when it is directly used for other instruments. This is because the instrumental variation causes the Local model's performance invalid. The instrumental variation is the difference among the spectra acquired by different instruments. The Global model identified the instrumental variation due to the Global model was developed with the calibration dataset acquired from two or more instruments. Thus, this study aims to compare the Local and Global models' performance among different NIR spectroscopy instruments in corn oil prediction. First, principle component analysis (PCA) was used to compress the NIR spectra. After that, Bayesian (BR) learning algorithm was applied to train an ANN with different initial conditions and hidden neurons to identify an optimal ANN for the primary instrument. The procedure of Global model development was similar to the Local model. The difference between Local and Global models is the global model used two or more calibration datasets to develop the model. Findings show that Global model was the best with the lowest root mean square error of prediction (RMSEP) of 0.1630%, followed by Local model of 0.4074% and 0.4330% for mp5 and mp6 as calibration datasets, respectively and the best correlation coefficient of 0.7514, 0.6532, and 0.7297, respectively, tested with m5 testing dataset in corn oils prediction applications. The same performance was found when the testing dataset were mp5 and mp6.

In response to the deficiencies in visible-near infrared spectral identification of coal gangue under different working conditions, this paper employs multiple algorithms to construct local and global recognition models and conducts analysis. Simulating five different working conditions in the laboratory, including different detection angles, illumination angles, detection distances, coal powder content, and moisture content. Each working condition is further divided into four levels. Collecting spectral data of coal and gangue samples. Constructing global and local datasets, which are respectively derived from all levels and single levels under certain working conditions. Training local and global models to identify samples based on machine learning algorithms such as Lightgbm and 1DCNN. The results of the experiment indicate that there are variations in the sample spectra and recognition accuracy based on different working conditions. Furthermore, the local model exhibits lower recognition accuracy and inadequate robustness for other level samples under the same working condition. The global model is more robust and practical, as it achieves higher or equal recognition accuracy for local data compared to the local model. Among the global models, Lightgbm has the best overall performance, with an average F1-score of 99.34% and 97.9% for local and global datasets, respectively. It also has fast training time of 0.75s and recognition time of 1 ms. The Lightgbm global model can accurately identify coal and gangue in various working conditions, making it a reliable solution.

Regional groundwater sequential forecasting using global and local LSTM models

A decentralized data evaluation framework in federated learning

This study aims to shed light on a new understanding of global machine-learning (ML) prediction models for wastewater treatment plants (WWTPs). The paper evaluates the development of local and global prediction models to predict the wastewater influent biochemical oxygen demand (BOD5) in four WWTPs. The paper proposes an integrated framework of remote sensing and ML techniques, specifically decision tree, random forest, adaptive boosting, and gradient boost algorithms. The modeling considered two cases for local and global models. The local model’s best score achieved 0.92 coefficient of determination (R2), 6.56 mean absolute error (MAE), and 1.00 R2, 0.08 MAE, respectively. In the second case, the global model’s best score achieved 0.58 R2, 20.73 MAE, and 0.95 R2, 3.93 MAE, respectively. The results showed that the developed local model reduced the BOD5 test results duration from five days to only three hours. However, in the first case, the developed global model failed to achieve good predictions against other WWTPs. This is due to the biological factors that change wastewater characteristics from one place to another. The study concluded that a local model is recommended to be developed for each WWTP separately. The novelty of this paper is that it investigates the various cases of testing the performance of the ML prediction models against different WWTPs than the one used to train and test. Also, it technically discusses the biological factors of wastewater that result in complications in ML modeling and prediction, which was never discussed or taken into consideration in ML literature.

SUMMARY We present a method to calculate local spherical harmonic power spectra directly from spacecraft magnetic and gravity data with varying spacecraft altitude. Previously published applications of local spherical harmonic power spectra have been formulated for data that are available at a single collection altitude, such as data evaluated from a global spherical harmonic model. Our approach consists of first solving for local models from local data and then obtaining local multitaper spectra from the local models. We demonstrate with numerical tests that this approach can produce reliable results. Our method is particularly useful in situations where data coverage does not allow for calculating global magnetic or gravity field models, or where data quality or quantity varies regionally and where local models could yield superior resolution or quality over global models.

Quantitative structure-activity relationships (QSARs) are broadly classified as global or local, depending on their molecular constitution. Global models use large and diverse training sets covering a wide range of chemical space. Local models focus on smaller structurally or chemically similar subsets that are conventionally selected by human experts or alternatively using clustering analysis. The current study focuses on the comparative analysis of different clustering algorithms (expectation-maximization, K-means and hierarchical) for seven different descriptor sets as structural characteristics and two rule-based approaches to select subsets for designing local QSAR models. A total of 111 local QSAR models are developed for predicting bioconcentration factor. Predictions from local models were compared with corresponding predictions from the global model. The comparison of coefficients of determination (r 2) and standard deviations for local models with similar subsets from the global model show improved prediction quality in 97% of cases. The descriptor content of derived QSARs is discussed and analyzed. Local QSAR models were further consolidated within the framework of consensus approach. All different consensus approaches increased performance over the global and local models. The consensus approach reduced the number of strongly deviating predictions by evening out prediction errors, which were produced by some local QSARs.

Federated learning is a newly emerging distributed learning framework that facilitates the collaborative training of a shared global model among distributed participants with their privacy preserved. However, federated learning systems are vulnerable to Byzantine attacks from malicious participants, who can upload carefully crafted local model updates to degrade the quality of the global model and even leave a backdoor. While this problem has received significant attention recently, current defensive schemes heavily rely on various assumptions, such as a fixed Byzantine model, availability of participants' local data, minority attackers, IID data distribution, etc. To relax those constraints, this paper presents Robust-FL, the first prediction-based Byzantine-robust federated learning scheme where none of the assumptions is leveraged. The core idea of the Robust-FL is exploiting historical global model to construct an estimator based on which the local models will be filtered through similarity detection. We then cluster local models to adaptively adjust the acceptable differences between the local models and the estimator such that Byzantine users can be identified. Extensive experiments over different datasets show that our approach achieves the following advantages simultaneously: (i) independence of participants' local data, (ii) tolerance of majority attackers, (iii) generalization to variable Byzantine model.

Electrocardiogram (ECG) data classification is a hot research area for its application in medical information processing. However, insufficient data, privacy preserve, and local deployment are still challenging difficulties. To address these problems, a novel personalized federated learning method for ECG classification is proposed in this paper. First, a global model is trained with federated learning framework on multiple local data clients. Then, we use the global model and private data to train the local model. To reduce the feature inconsistency between global and private local data and for better fitting the private local data, a novel ”feature alignment” module is devised to guarantee the uniformity, which contains two parts, global alignment and local alignment, respectively. For global alignment, the graph metric of batch data is used to constrain the dissimilarity between features generated by the global model and local model. For local alignment, triplet loss is adopted to increase discriminative ability for local private data. Comprehensive experiments on our collected dataset are evaluated. The results show that the proposed method can be better adapted to local data and exhibit superior ability of generalization.

Purpose. Construction of a global model of an aircraft landing gear compartment, study of the stress-strain state of the compartment structure zones, and a comparative analysis of it with a local model. The main task of this work is to assess the possibility of using a global model for further assessment of the resource and static strength of the structure of the aircraft landing gear compartment. Research methods. A global model was developed and numerical calculations of the stress-strain state of the nose strut compartment of the aircraft landing gear were obtained using the PATRAN/NASTRAN software complex. Quad4 type elements were used to create a mesh of flat elements. The Mesh Seed tool was used to obtain a mesh of satisfactory dimensions. The mesh creation method is Iso Mesh. All beam elements of the structure were modeled using Bar elements. For the numerical solution of the presented problem, the nonlinear solver 106 - Nonlinear Static was used. Results. A numerical calculation of the stress-strain state of the nose strut of the aircraft landing gear during global modeling was carried out. Visualization of linear movement, normal stresses on the upper and lower surface of the wall, deflection of the wall and the plate between the reinforcement elements, local representation of the equilibrium of the element in terms of axial force and bending moment was obtained. Graphs of these characteristics of the stress-strain state of the structure are constructed. A comparative analysis of the results of calculations of local and global models was made. Scientific novelty. Application of the method of calculating the stress-strain state of complex structures of aviation equipment elements using computer software complexes in the conditions of modern development of aviation science and technology. Practical value. The results of the comparative analysis of global and local models and the method of numerical calculation of the stress-strain state of a complex structure can be used for further assessment of the resource and static strength of the structure of the aircraft landing gear compartment.

Personalized federated learning-based intrusion detection system: Poisoning attack and defense

32 Quantitative Structure-Property Relationship (QSPR) models were constructed for prediction of aqueous intrinsic solubility of liquid and crystalline chemicals. Data sets contained 1022 liquid and 2615 crystalline compounds. Multiple Linear Regression (MLR), Support Vector Machine (SVM) and Random Forest (RF) methods were used to construct global models, and k-nearest neighbour (kNN), Arithmetic Mean Property (AMP) and Local Regression Property (LoReP) were used to construct local models. A set of the best QSPR models was obtained: for liquid chemicals with RMSE (root mean square error) of prediction in the range 0.50-0.60 log unit; for crystalline chemicals 0.80-0.90 log unit. In the case of global models the large number of descriptors makes mechanistic interpretation difficult. The local models use only one or two descriptors, so that a medicinal chemist working with sets of structurally-related chemicals can readily estimate their solubility. However, construction of stable local models requires the presence of closely related neighbours for each chemical considered. It is probable that a consensus of global and local QSPR models will be the optimal approach for construction of stable predictive QSPR models with mechanistic interpretation.

The finite element method has been widely used to enable fatigue life predictions for electronic assemblies subjected to thermal cycling. The potential critical region of interest within a solder joint is relatively small relative to the entire package assembly. High-density meshes are typically used to build a package model when using traditional FE modeling approaches. Therefore, long computational times can be expected for an analysis involving several thermal cycles. In order to reduce the complexity of the model and the use of oversize elements, and improve the efficiency of calculation, the technique of global/local modeling (submodeling) has been developed. In this approach, interpolated displacements from coarse model (global model) are applied as boundary conditions to a refined model (local/submodel) of the critical area of interest. The accuracy and efficiency of submodeling finite element simulations for electronic packages have not been evaluated completely in the literature.In this work, submodeling approaches for BGA assemblies have been explored in detail. A typical BGA package assembly was modeled using several coarse meshes as global models; and the maximum aspect ratio within the global model was varied up to 25, which is near the warning limit of many commercial finite element codes. Once the critical solder joint was identified, then the local model was built using the technique of cut boundaries, and meshed with refined elements. The maximum aspect ratio was also reduced to 5. The simulation result shows the accuracy of solution was sensitive to the mesh quality of the local model, as well as the load step size for both global and local models. An improved simulation strategy using submodeling was developed to obtain the best compromise in the global and local models between the mesh quality and load step size.Although the submodeling is a powerful tool for a better solution in a local region of interest far away from the cut boundaries, it is still crucial to identify the critical region correctly from the global model or a misguided local solution might be performed. Initially, the global models were built using coarse meshes with nonlinear material properties as seen in literature, and then an improved geometric simplification of the solder joint incorporating energy based fatigue criteria was developed. In addition, minimization of the local model volume is still required for a detail analysis of a complex configuration in local region where a high-density mesh is needed. In the investigated approach, we determined the minimized height of the local boundary, and have shown the influence between the height and the solution of volume averaged inelastic strain energy dissipation in the local model. The proposed approach achieves a large reduction in computational time, better detailed modeling of local interest region, and improved simulation accuracy.

In federated learning (FL), a server determines a global learning model by aggregating the local learning models of clients, and the determined global model is broadcast to all the clients. However, the global learning model can significantly deteriorate if a Byzantine attacker transmits malicious learning models trained with incorrectly labeled data. We propose a Byzantine-robust FL algorithm that, by employing a consensus confirmation method, can reduce the success probability of Byzantine attacks. After aggregating the local models from clients, the proposed FL server validates the global model candidate by sending the global model candidate to a set of randomly selected FL clients and asking them to perform local validation with their local data. If most of the validation is positive, the global model is confirmed and broadcast to all the clients. We compare the performance of the proposed FL against Byzantine attacks with that of existing FL algorithms analytically and empirically.

Machine learning (ML) has become an indispensable tool to predict absorption, distribution, metabolism, and excretion (ADME) properties in pharmaceutical research. ML algorithms are trained on molecular structures and corresponding ADME assay data to develop quantitative structure-property relationship (QSPR) models. Traditional QSPR models were trained on compound sets of limited size. With the advent of more complex ML algorithms and data availability, training sets have become larger and more diverse. Most common training approaches consist in either training a model with a small set of similar compounds, namely, compounds designed for the same drug discovery project or chemical series (local model approach) or with a larger set of diverse compounds (global model approach). Global models are built with all experimental data available for an assay, combining compound data from different projects and disease areas. Despite the ML progress made so far, the choice of the appropriate data composition for building ML models is still unclear. Herein, a systematic evaluation of local and global ML models was performed for 10 different experimental assays and 112 drug discovery projects. Results show a consistent superior performance of global models for ADME property predictions. Diagnostic analyses were also carried out to investigate the influence of training set size, structural diversity, and data shift in the relative performance of local and global ML models. Training set and structural diversity did not have an impact in the relative performance on the methods. Instead, data shift helped to identify the projects with larger performance differences between local and global models. Results presented in this work can be leveraged to improve ML-based ADME properties predictions and thus decision-making in drug discovery projects.