FL Model Research Articles

Automatic Gesture-Based Arabic Sign Language Recognition: A Federated Learning Approach

Featuring machine learning algorithms for recognizing hand gesture patterns adjusted for individuals with disabilities is an expanding trend in assisted living. This paper addresses the challenge of interpreting the semantics of image-based hand gestures by introducing a federated deep learning architecture for Arabic sign language recognition. The proposed model manages distributed learning through a client-server paradigm, wherein several edge nodes collaborate to jointly learn the discriminative features of confidential data without breaching its privacy. This model will enable more accessibility for people with deafness or impairment using image gestures. The federated learning procedure is primarily based on the ResNet32 deep backbone and federated averaging mechanism. The experimental results show the effectiveness of the proposed FL model, achieving an accuracy of 98.30% with 33 seconds on average for each client in a single training round. This demonstrates its high capabilities in recognizing Arabic sign language and improving the communication experience for people with disabilities.

A Game-theoretic Framework for Privacy-preserving Federated Learning

In federated learning, benign participants aim to optimize a global model collaboratively. However, the risk of privacy leakage cannot be ignored in the presence of semi-honest adversaries. Existing research has focused either on designing protection mechanisms or on inventing attacking mechanisms. While the battle between defenders and attackers seems never-ending, we are concerned with one critical question: Is it possible to prevent potential attacks in advance? To address this, we propose the first game-theoretic framework that considers both FL defenders and attackers in terms of their respective payoffs, which include computational costs, FL model utilities, and privacy leakage risks. We name this game the federated learning privacy game (FLPG), in which neither defenders nor attackers are aware of all participants’ payoffs. To handle the incomplete information inherent in this situation, we propose associating the FLPG with an oracle that has two primary responsibilities. First, the oracle provides lower and upper bounds of the payoffs for the players. Second, the oracle acts as a correlation device, privately providing suggested actions to each player. With this novel framework, we analyze the optimal strategies of defenders and attackers. Furthermore, we derive and demonstrate conditions under which the attacker, as a rational decision-maker, should always follow the oracle’s suggestion not to attack .

A hybrid fuzzy logic/genetic algorithm model based on experimental data for estimation of cuttings concentration during drilling

Adequate hole cleaning is a key to economic drilling operations. However, cuttings transport during drilling is affected by different operational parameters, geometric parameters and, cuttings and drilling fluid related parameters which makes the cuttings transport phenomenon a complicated modeling exercise. This complexity has led to the use of computational intelligence (CI) methods such as fuzzy logic and genetic algorithm (GA). A total of 11 parameters are considered in the current work while developing Mamdani type fuzzy logic model using 702 experimental observations. The developed FL model (FL2) is further combined with GA to optimize fuzzy rule weights which has led to the development of a hybrid CI model termed FL3 in this paper. The performance of FL3 over two test datasets is compared with FL2 and a previously published FL model (Chowdhury and Hovda, 2022) termed FL1 (built on 509 experimental observations and 195 fuzzy rules). Model comparison using three goodness of fit metrics (R2, RMSE and MAE) shows that the hybrid CI model, FL3, outperforms both FL1 and FL2 over both test datasets. R2 value of FL3 over both test datasets is ‘very good’ following (Moriasi et al., 2007). The difference of FL3 cuttings concentration (Cc) estimates and the measured Cc values is less than 1% for more than two thirds of the experimental data for both test datasets collected at two different geographical locations by two different research groups – TUDRP and SINTEF. The reasons behind FL3 outperforming FL1 and FL2 are data randomization, modified fuzzy sets for eccentricity and drillstring rotation based on the exploratory data analysis published earlier (Chowdhury et al., 2023), use of more experimental data and the optimization of fuzzy rule weights by GA.

Application of Mamdani Fuzzy Inference System in Poultry Weight Estimation.

Traditional manual weighing systems for birds on poultry farms are time-consuming and may compromise animal welfare. Although automatic weighing systems have been introduced as an alternative, they face limitations in accurately estimating the weight of heavy birds. Therefore, exploring alternative methods that offer improved efficiency and precision is necessary. One promising solution lies in the application of AI, which has the potential to revolutionize various aspects of poultry production and management, making it an indispensable tool for the modern poultry industry. This study aimed to develop an AI approach based on the FL model as a viable solution for estimating poultry weight. By incorporating expert knowledge and considering key input variables such as indoor temperature, indoor humidity, and feed consumption, FL-based models were developed with different configurations using Mamdani inferences and evaluated across eight different rearing periods in Samsun, Türkiye. This study's results demonstrated the effectiveness of FL-based models in estimating poultry weight. The models achieved varying average absolute error values across different age groups of broilers, ranging from 0.02% to 5.81%. These findings suggest that FL-based methods hold promise for accurate and efficient poultry weight estimation. This study opens up avenues for further research in the field, encouraging the exploration of FL-based approaches for improved poultry weight estimation in poultry farming operations.

Flipped learning as an educational model in a cardiology residency program

IntroductionFlipped learning (FL) is a model which allows students to adjust their study rhythm by taking advantage of class time to apply their knowledge. Although FL meets many of the needs of the traditional lecture-based model and the exclusively virtual model, its effectiveness in medical residency programs has not been thoroughly studied. Our study evaluates the efficacy of an FL model in a cardiology residency program based on the satisfaction and academic performance of the participants.MethodA prospective, observational, and descriptive study was conducted at Ignacio Chávez National Institute of Cardiology in Mexico City to evaluate the satisfaction and effectiveness of an FL model for acquiring competencies in electrocardiographic diagnosis among thirty-one first-year cardiology residents. The Advanced Electrocardiography Workshop and a virtual classroom were designed for the intervention. Four teachers taught the workshop and video classes, and ten medical specialists from the participants’ areas of work performed evaluations of practical skills before and after the intervention.Results75% of the participants rated the intervention as very favorable compared to the traditional and virtual models. The main advantages offered by the FL model were identified. An improvement of high statistical significance was observed in participants’ academic performance after the intervention (P value < 0.001).ConclusionThe FL model has a high degree of acceptance among participants and teachers. Our study shows an improvement in academic performance after the intervention. Considering that the FL model is flexible and reproducible in other areas of medicine, it represents an excellent educational alternative that meets current needs.

Simulating a flexible water model as rigid: Best practices and lessons learned.

Two ways to create rigid versions of flexible models are explored. The rigid model can assume the Model's Geometry (MG) as if the molecule is not interacting with any other molecules or the ensemble averaged geometry (EG) under a particular thermodynamic condition. Although the MG model is more straightforward to create, it leads to relatively poor performance. The EG model behaves similarly to the corresponding flexible model (the FL model) and, in some cases, agrees even better with experiments. While the difference between the EG and the FL models is mostly a result of flexibility, the MG and EG models have different dipole moments as a result of an effective induction in the condensed phase. For the three water models studied, the property that shows the most difference is the temperature dependence of density. The MG version of the water model by adaptive force matching for ice and liquid does not possess a temperature of maximum density, which is attributed to a downshift of the putative liquid-liquid phase transition line, leading to the hypothesized second critical point of liquid water to manifest at negative pressure. A new three-phase coexistence method for determining the melting temperature of ice is also presented.

A group key exchange and secure data sharing based on privacy protection for federated learning in edge‐cloud collaborative computing environment

SummaryFederated learning (FL) is widely used in internet of things (IoT) scenarios such as health research, automotive autopilot, and smart home systems. In the process of model training of FL, each round of model training requires rigorous decryption training and encryption uploading steps. The efficiency of FL is seriously affected by frequent encryption and decryption operations. A scheme of key computation and key management with high efficiency is urgently needed. Therefore, we propose a group key agreement technique to keep private information and confidential data from being leaked, which is used to encrypt and decrypt the transmitted data among IoT terminals. The key agreement scheme includes hidden attribute authentication, multipolicy access, and ciphertext storage. Key agreement is designed with edge‐cloud collaborative network architecture. Firstly, the terminal generates its own public and private keys through the key algorithm then confirms the authenticity and mapping relationship of its private and public keys to the cloud server. Secondly, IoT terminals can confirm their cryptographic attributes to the cloud and obtain the permissions corresponding to each attribute by encrypting the attributes. The terminal uses these permissions to encrypt the FL model parameters and uploads the secret parameters to the edge server. Through the storage of the edge server, these ciphertext decryption parameters are shared with the other terminal models of FL. Finally, other terminal models are trained by downloading and decrypting the shared model parameters for the purpose of FL. The performance analysis shows that this model has a better performance in computational complexity and computational time compared with the cited literature.

Federated Unlearning: Guarantee the Right of Clients to Forget

The Right to be Forgotten gives a data owner the right to revoke their data from an entity storing it. In the context of federated learning, the Right to be Forgotten requires that, in addition to the data itself, any influence of the data on the FL model must disappear, a process we call “federated unlearning.” The most straightforward and legitimate way to implement federated unlearning is to remove the revoked data and retrain the FL model from scratch. Yet the computational and time overhead associated with fully retraining FL models can be prohibitively expensive. In this article, we take the first step to comprehensively investigate the way to settle the unlearning paradigm in the context of federated learning. First, we define the problem of efficient federated unlearning, including its challenges and goals, and we identify three common types of federated unlearning requests: class unlearning, client unlearning, and sample unlearning. Based on those challenges and goals, a general pipeline is proposed for federated unlearning for the above three types of requests. We revisit how the training data affects the final FL model performance and thereby empowers the proposed framework with the reverse stochastic gradient ascent (SGA) and elastic weight consolidation (EWC). Various experiments are conducted to verify effectiveness of the proposed method in both aspects of unlearning efficacy and efficiency. We believe the proposed method will perform as an essential component for future machine unlearning systems.

Communication-Efficient Federated Edge Learning for NR-U-Based IIoT Networks

As a key infrastructural technology, Industrial Internet of Things (IIoT) and its related techniques have emerged in the age of Industrial Internet. Among them, an increasing popular and attractive federated edge learning (FEL) mechanism, which performs data analysis and inference at the edge devices distributedly, and aggregates local FEL units at a centralized controller, is introduced to meet the stringent data privacy and low-latency requirements for high-stake IIoT devices. Due to the bandwidth limitation, only parts of the IIoT devices can be selected to transmit their local FEL models to the centralized controller at each learning step. However, the centralized controller prefers to collect all the local FEL models to generate the global FL model since each IIoT device has a differential data set. Existing works mainly focus on selecting an appropriate subset of IIoT devices through advanced scheduling mechanisms without extending the resource bandwidth. However, the new radio in unlicensed spectrum (NR-U) technology in the 5G network opens up new possibilities for FEL since it is a privately owned network with fruitful bandwidth resources. We thus propose a novel communication-efficient FEL mechanism for NR-U-based IIoT networks, which aims to select data importance IIoT devices for local training under relatively sufficient unlicensed resources. The objective function is formulated as a tradeoff between total FEL data importance and the transmission latency via joint learning, device selection, and resource management scheduling, which is a mixed-integer nonlinear programming (MINLP). To deal with this problem, an alternating direction method of multipliers and block coordinate update (ADMM-BCU) algorithm with low computational complexity has been used. Closed-form expressions for both optimal device selection and resource management are derived, which highlighted significant insights. Numerical results demonstrate the algorithmic advantages and structural benefits of the proposed strategies.

Privacy-preserving Speech-based Depression Diagnosis via Federated Learning.

Mental health disorders, such as depression, affect a large and growing number of populations worldwide, and they may cause severe emotional, behavioral and physical health problems if left untreated. As depression affects a patient's speech characteristics, recent studies have proposed to leverage deep-learning-powered speech analysis models for depression diagnosis, which often require centralized learning on the collected voice data. However, this centralized training requiring data to be stored at a server raises the risks of severe voice data breaches, and people may not be willing to share their speech data with third parties due to privacy concerns. To address these issues, in this paper, we demonstrate for the first time that speech-based depression diagnosis models can be trained in a privacy-preserving way using federated learning, which enables collaborative model training while keeping the private speech data decentralized on clients' devices. To ensure the model's robustness under attacks, we also integrate different FL defenses into the system, such as norm bounding, differential privacy, and secure aggregation mechanisms. Extensive experiments under various FL settings on the DAIC-WOZ dataset show that our FL model can achieve high performance without sacrificing much utility compared with centralized-learning approaches while ensuring users' speech data privacy. Clinical Relevance- The experiments were conducted on publicly available clinical datasets. No humans or animals were involved.

Assessment and application of an alternative formula to describe the hydraulic conductivity over the full moisture range

AbstractMotivated by a versatile three‐parameter water retention curve (WRC) formula of Fredlund and Xing, we proposed an alternative expression to describe the relative hydraulic conductivity (RHC) for soils that can capture oven‐dry to water‐saturated conditions (full moisture range) with a single, closed‐form equation (without having to explicitly partitioning the capillary and film flow components). This RHC expression was tested using an existing data set that includes a wide range of soil textures, and shown to be more accurate than the other two RHC expressions. Subsequently, the new RHC expression was combined with the WRC formula of Fredlund and Xing, generating an alternative model (hereafter called the FL model) applicable to describe soil hydraulic properties over the full moisture range. The FL model requires both the known WRC and RHC data since three additional fitting parameters are involved in the present RHC expression. Furthermore, the FL model was employed in a HYDRUS‐1D numerical model for simulating the soil moisture redistribution within a lysimeter. The numerical simulation with using the FL model can accurately predict the moisture redistribution, in turn confirming the validity of the FL model despite without partitioning the capillary and film flow components.

GRNN: Generative Regression Neural Network—A Data Leakage Attack for Federated Learning

Data privacy has become an increasingly important issue in Machine Learning (ML) , where many approaches have been developed to tackle this challenge, e.g., cryptography ( Homomorphic Encryption (HE) , Differential Privacy (DP) ) and collaborative training (Secure Multi-Party Computation (MPC) , Distributed Learning, and Federated Learning (FL) ). These techniques have a particular focus on data encryption or secure local computation. They transfer the intermediate information to the third party to compute the final result. Gradient exchanging is commonly considered to be a secure way of training a robust model collaboratively in Deep Learning (DL) . However, recent researches have demonstrated that sensitive information can be recovered from the shared gradient. Generative Adversarial Network (GAN) , in particular, has shown to be effective in recovering such information. However, GAN based techniques require additional information, such as class labels that are generally unavailable for privacy-preserved learning. In this article, we show that, in the FL system, image-based privacy data can be easily recovered in full from the shared gradient only via our proposed Generative Regression Neural Network (GRNN) . We formulate the attack to be a regression problem and optimize two branches of the generative model by minimizing the distance between gradients. We evaluate our method on several image classification tasks. The results illustrate that our proposed GRNN outperforms state-of-the-art methods with better stability, stronger robustness, and higher accuracy. It also has no convergence requirement to the global FL model. Moreover, we demonstrate information leakage using face re-identification. Some defense strategies are also discussed in this work.

Reconfigurable Intelligent Surface Enabled Federated Learning: A Unified Communication-Learning Design Approach

To exploit massive amounts of data generated at mobile edge networks, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">federated learning</i> (FL) has been proposed as an attractive substitute for centralized machine learning (ML). By collaboratively training a shared learning model at edge devices, FL avoids direct data transmission and thus overcomes high communication latency and privacy issues as compared to centralized ML. To improve the communication efficiency in FL model aggregation, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">over-the-air computation</i> has been introduced to support a large number of simultaneous local model uploading by exploiting the inherent superposition property of wireless channels. However, due to the heterogeneity of communication capacities among edge devices, over-the-air FL suffers from the straggler issue in which the device with the weakest channel acts as a bottleneck of the model aggregation performance. This issue can be alleviated by device selection to some extent, but the latter still suffers from a tradeoff between data exploitation and model communication. In this paper, we leverage the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">reconfigurable intelligent surface</i> (RIS) technology to relieve the straggler issue in over-the-air FL. Specifically, we develop a learning analysis framework to quantitatively characterize the impact of device selection and model aggregation error on the convergence of over-the-air FL. Then, we formulate a unified communication-learning optimization problem to jointly optimize device selection, over-the-air transceiver design, and RIS configuration. Numerical experiments show that the proposed design achieves substantial learning accuracy improvement compared with the state-of-the-art approaches, especially when channel conditions vary dramatically across edge devices.

Augmented Reality Applications in Science Experiment Practices

The purpose of this paper is to explore the impact of augmented reality (AG) and the flipped learning model on the attitude of 4th-grade students towards the science and technology course and to determine the opinion of students concerning videos presented through AG. Another objective of the study is to determine parents' opinions concerning AG and FL in science and technology courses. A pre-test / post-test quasi-experimental design was used with the control group to obtain quantitative data. Qualitative data were also included in the analysis to support quantitative data obtained from a case study pattern. In the study, the Science and Technology Attitude Scale was used to measure students' attitudes. The Opinion Scale for Experimental Videos was used to evaluate video usage opinions in science and technology courses. Besides, parents' opinions concerning AG applications and the FL model were collected through the researchers' interview form. Independent sample t-test and ANCOVA tests were conducted to interpret the quantitative data of the study. Qualitative data were evaluated with the content analysis method. The results showed that the AR and FL model's application provides students with a positive attitude towards science and technology; it also helped them solve problems. Besides, families have also expressed positive opinions about AR and FL applications.

Fuzzy logic to thermal and friction characteristics of turbulent air-flow over diamond-shaped turbulators

The research tries to apply a fuzzy logic approach to approximate the airflow thermal characteristic and flow friction for the turbulent regime in a heat exchanger equipped by diamond-shaped (D-shape) turbulators with arrangements in tandem. The variables studied in the work are Reynolds number within the range of 3500 to 16,500, three cone angles (θ), and three tail length ratios (TR). The heat transfer and pressure loss increased by increasing and decreasing the cone angle and tail length ratio, respectively. Investigation of the fuzzy logic model indicated that it is a powerful method to estimate the thermal characteristic and flow friction in the system. The accuracy of the fuzzy model was investigated and the mean relative errors (MREs) of 0.61% and 0.79% were calculated for predicting the Nusselt number and friction factor, respectively. In this study, the previous empirical correlations are corrected and new correlations are presented with higher accuracy. The constants of the equations were modified using the genetic algorithm. The prediction accuracy of the empirical correlations was increased especially for the Nusselt number. The errors of the FL model were compared with empirical correlations, which indicated the superiority of the fuzzy model.

A prompt CNN-FL model based on multi criteria for a smart vertical handoff in heterogeneous system

PurposeHeterogeneous network is mainly focused to enrich the demands of network's traffic and data rate. Heterogeneous network is an integrated system composed of diverse radio access technologies such as Wi-Fi, UMTS and WiMAX soon. To exchange the packets among these different wireless technologies, an adoptable vertical handoff (VHO) is required.Design/methodology/approachVarious types of techniques have been proposed in line with VHO. However these algorithms lead to the threat of high dropping rate and poor Quality of Service (QoS). We modulate a new methodology by integrating Convolutional Neural Network with Fuzzy Logic (CNN-FL) for an optimum vertical handoff decision process (VHDP). The CNN-FL–based VHDP is very effective because of nonlinearity and generalization capability with uncertain data inputs.FindingsThe performance results exhibit that proposed methods show efficient for selecting the appropriate network with maximum throughput. At the same time, the proposed method reduces computational complexity and dropping rate considerably.Originality/valueThis paper proposes a novel CNN model along with a FL model for achieving a vertical hand off.

Flipped Learning in Information Technology Courses: Effectiveness and Impact

This study’s purpose is to investigate the implication of the use of flipped learning on students’ performance and perception. Participants are the second-year and third-year undergraduate students enrolled in the Information Technology courses. One hundred and twenty-five students were the participants associate with an experimental group and a control group. The experimental group consisted of the students of the ISD class A and BI courses. Meanwhile, the students of ISD class D and MIT were the control group. The experiments employed two instruments: (1) students’ final score, and (2) questionnaire. This study used the independent sample t-test and two-proportion z-test to analyze the collected data. The result showed that the students’ performance in the FL classes was superior to a traditional learning setting. Students’ perception of both courses is higher than the experimental group. Besides, we observed that studying in an FL model had a positive impact on students’ motivation. These findings suggest that the FL could be a promising way of enhancing students’ learning outcomes.

Federated Learning With Differential Privacy: Algorithms and Performance Analysis

In this paper, to effectively prevent information leakage, we propose a novel framework based on the concept of differential privacy (DP), in which artificial noises are added to the parameters at the clients side before aggregating, namely, noising before model aggregation FL (NbAFL). First, we prove that the NbAFL can satisfy DP under distinct protection levels by properly adapting different variances of artificial noises. Then we develop a theoretical convergence bound of the loss function of the trained FL model in the NbAFL. Specifically, the theoretical bound reveals the following three key properties: 1) There is a tradeoff between the convergence performance and privacy protection levels, i.e., a better convergence performance leads to a lower protection level; 2) Given a fixed privacy protection level, increasing the number $N$ of overall clients participating in FL can improve the convergence performance; 3) There is an optimal number of maximum aggregation times (communication rounds) in terms of convergence performance for a given protection level. Furthermore, we propose a $K$-random scheduling strategy, where $K$ ($1<K<N$) clients are randomly selected from the $N$ overall clients to participate in each aggregation. We also develop the corresponding convergence bound of the loss function in this case and the $K$-random scheduling strategy can also retain the above three properties. Moreover, we find that there is an optimal $K$ that achieves the best convergence performance at a fixed privacy level. Evaluations demonstrate that our theoretical results are consistent with simulations, thereby facilitating the designs on various privacy-preserving FL algorithms with different tradeoff requirements on convergence performance and privacy levels.

Modeling of Compressive Strength of Concrete using Gaussian Membership Function

This paper presents an application of fuzzy logic to forecast the compressive strength of concrete. The fuzzy model examines 7 different input parameters that comprises: Cement, Coarse aggregate(CA), Super plasticizer(SP), Fine Aggregate(FA), Slag, Fly ash, Water(W), and 28 days compressive strength is taken as the output parameter. By using Gaussian membership function, the fuzzy logic technique is used for developing models. For assessing the results of FL model with experimental results, root mean square error, mean absolute error and correlation coefficient are used. The results showed that FL can be a better modeling tool and an another technique for predicting the concrete’s compressive strength.

Rethinking the flipped learning pre‐class: Its influence on the success of flipped learning and related factors

AbstractFlipped learning (FL) has had a substantial impact on higher education worldwide, prompting active dialog among educational researchers and practitioners. It is widely accepted that the value of FL lies in enabling in‐depth, learner‐centered learning experiences by offloading content delivery onto pre‐class online learning. The in‐class component is thus generally considered more critical, even though both pre‐class and in‐class elements are indispensable components of FL and practitioners are faced with the issue of under‐prepared learners. In this study, we examine the significance of pre‐class learning as a gateway to the intended benefits of FL. We aim to investigate the extent to which FL pre‐class learning influences overall FL success and what other factors affect FL pre‐class learning to ascertain the implications of designing and implementing FL pre‐class learning. A college life science course was designed and implemented with the FL model for a semester, from which data were collected and statistically analyzed. The results indicate that the influence of pre‐class learning on final FL success is significant and strong. Regarding the modality of pre‐class learning materials, learners demonstrated higher levels of performance in the weeks with video materials than during those with reading materials. Among several learner factors, self‐direction was found to significantly influence pre‐class learning performance and perceived learning readiness. The role of FL pre‐class learning to attain the intended value of FL is discussed, and theoretical and practical design implications are suggested.