Malicious Workers Research Articles

Reliable federated learning based on dual-reputation reverse auction mechanism in Internet of Things

Federated learning, a promising distributed machine learning paradigm, has been used in various Internet of Things (IoT) environments to solve isolated data island issues and protect data privacy. However, since the central server in federated learning cannot detect the local training process of the client, it is vulnerable to adversarial attacks against its security and privacy by malicious clients during the training process. To address this problem, this work proposes a federated learning system model based on dual-reputation reverse auction in IoT. Specifically, due to resource constraints, not all clients participate in the federated learning process, so we propose the reputation-bid ratio based greedy half-splitting algorithm to select some clients to participate in the federated learning, which can guarantee that each client has the chance to be selected while selecting as many honest and high-quality clients as possible. Then, we propose an adaptive dropout aggregation method based on a training quality score, which can effectively defend against malicious workers’ attacks. After the completion of federated learning, we put forward a subjective evaluation incentive mechanism based on a second reputation to realize the fair incentive. Furthermore, we store and manage reputations through blockchain in our model to ensure their availability. Theoretical analysis deduces the complexity and security of our proposed model. Finally, simulation results indicate that our proposed model can achieve the highest accuracy across all test datasets compared to the baseline. Particularly, on the CIFAR10 dataset, the accuracy of our proposed model surpasses the baseline by 5% to 30%. In responding to sudden attacks initiated by normally participating workers, our model exhibits the fastest reaction time, with accuracy surpassing the baseline by 8% to 40%.

DTC-MDD: A spatiotemporal data acquisition technology for privacy-preserving in MCS

Some attackers in the Internet of Things submit falsified high-quality data to cause harm to users. To prevent malicious workers from reporting untruthful data for skyline computation in Mobile Crowd Sensing, we propose a double trust check-based spatiotemporal data acquisition scheme, DTC-MDD. In DTC-MDD, worker trust uses four-way validation to obtain reliable worker trust evaluations. Then, based on Probabilistic Skyline Calculation, we propose a worker selection algorithm to select high-trust, high-quality workers for data reporting. We also introduce the Non-Interactive Encrypted Integer Comparison Protocol to safeguard privacy between workers and users from malicious attacks. Finally, through extensive simulations on real datasets, DTC-MDD effectively enhances the quality and security of spatiotemporal data acquisition. DTC-MDD improved the data quality and reliability of candidate worker sets by 16.2% and 49.1%, respectively, and the data quality and reliability of the first skyline worker by 21.4% and 320.0%, respectively.

Collusion Detection and Trust Management for Indoor Navigation System With Crowdsourcing

The indoor navigation system supported by spatial crowdsourcing emerges as a promising application to provide customized location service for requesters. An important stage of crowdsourcing is to select trustworthy workers. Workers' reputation, as an essential criterion of this selection, is usually evaluated by feedback ratings from requesters. However, the reputation in the crowdsourcing-based indoor navigation system is vulnerable to the collusion attack, that is malicious workers (i.e., attackers) collude with requesters to illegally increase reputation. In this paper, we propose a collusion detection scheme to distinguish attackers and provide a secure reputation mechanism. Specifically, we first identify collusive requesters categorized into three different levels according to their feedback rating behaviors. Then, the weighted logistic regression (WLR) is developed to distinguish the collusive requesters who provide exorbitant feedback ratings. Furthermore, we employ an outlying sequence detection based on the maximum mean discrepancy (MMD), to resist the multiple location queries initiated by the same collusive requester through analyzing the distribution distance. In addition, we propose a community detection algorithm, named Fastgreedy, to identify the collusion from many requesters. Finally, the extensive simulation results demonstrate that the proposed scheme can effectively detect collusive requesters and significantly outperform other methods.

RLTD: A Reinforcement Learning-based Truth Data Discovery scheme for decision support systems under sustainable environments

The online world and associated information and communication technologies have generated digital networks by processing massive volumes of data and have a significant impact on the environmental sustainability. Mobile Crowd Source (MCS) is one of the digital technologies that can help humanity to better sense, understand and protect the environment by using vast amounts of data obtained to construct intelligent decision support systems (DSS). However, as the false data submitted by dishonest and malicious workers will cause the data-based DSSs to make wrong decisions and thus cause great harm, it is an urgent issue to propose an effective Truth Data Discovery (TDD) scheme for MCS. To tackle this issue, a Reinforcement Learning-based Truth Data Discovery (RLTD) scheme is proposed to obtain truth data in MCS at low cost in this paper. The main innovations of the RLTD scheme are as follows: (1) A novel trustworthiness-based TDD scheme is proposed to obtain truth data accurately at low cost, which can facilitate data-based DSSs in MCS. (2) Combined with Matrix Factorization (MF), we propose a worker recruitment method that only needs to recruit ϑn (ϑ≤1) workers for TDD in n tasks, which reduces the data collection cost significantly than previous TDD schemes. (3) We propose a Reinforcement Learning-based Site Selection (RLSS) method that intelligently selects as few sites as possible for worker recruitment with guaranteed high data quality. Experimental results demonstrate that the RLTD scheme can improve the accuracy of data collection by 1.31%–21.02%, reduce the data collection cost by 81.39%–85.50% compared to the traditional TDD schemes, and identify workers with accuracy of 86.67%–94.58%.

Pub-SubMCS: A privacy-preserving publish–subscribe and blockchain-based mobile crowdsensing framework

This paper proposes a privacy-preserving publish–subscribe-based decentralized framework for MCS systems named “Pub-SubMCS”. The framework allows data sharing, where requesters can subscribe to an existing data request (task) if their requirements match. Otherwise, they can create a new task with specific requirements on considered parameters. Incorporating the publish–subscribe (pub–sub) service model in a decentralized MCS system saves system entities’ sensing and computing resources and the cost of acquiring the data by the requesters. However, the pub–sub service model makes the curse of sensing issues more severe. Pub-SubMCS handles the curse of sensing issues by performing access control using smart contracts, which impose restrictions on data collectors (workers) to publish the data and identify and penalize the malicious workers early. To ensure data privacy and validation simultaneously over blockchain, we perform data transformation enabling the validation algorithm to run over transformed data and thus enhancing trust among the system entities. In particular, we use the normalization technique to transform data and the Pearson correlation coefficient measure to compare the similarity in the collected sensor data. Pub-SubMCS is implemented on the Ethereum blockchain, and solidity programming language is used to create smart contracts. The security analysis and experiment results show the proposed system’s scalability, usability, and feasibility. We also demonstrate the effectiveness of the publish–subscribe model against the requester–worker model.

Byzantine Tolerant Algorithms for Federated Learning

In federated learning, workers periodically upload locally computed weights to a federated learning server (FL server). When Byzantine attacks are presented in the system, attacked workers may upload incorrect weights to the parameter server, i.e., the information received by the FL server is not always the true values computed by workers. Previously proposed score-based, median-based, and distance-based defense algorithms made the following assumptions unrealistic in federated learning: (1) the dataset on each worker is independent and identically distributed (i.i.d.), and (2) the majority of all participating workers are honest. In federated learning, however, a worker may keep its non-i.i.d. private dataset and malicious workers may take over the majority in some iterations. In this paper, we focus on model poisoning type Byzantine attack and propose a novel reference dataset based algorithm along with a practical Two-Filter algorithm (ToFi) to defend against Byzantine attacks in federated learning. Our experiments highlight the effectiveness of our algorithm compared with previous algorithms in different settings.

Reliable Distributed Computing for Metaverse: A Hierarchical Game-Theoretic Approach

The metaverse is regarded as a new wave of technological transformation that provides a virtual space for people to interact through digital avatars. To achieve immersive user experiences in the metaverse, real-time rendering is the key technology. However, computing intensive tasks of real-time rendering from metaverse service providers cannot be processed efficiently on a single resource-limited mobile device. Alternatively, such mobile devices can offload the metaverse rendering tasks to other mobile devices by adopting the collaborative computing paradigm based on Coded Distributed Computing (CDC). Therefore, this paper introduces a hierarchical game-theoretic CDC framework for the metaverse services, especially for vehicular metaverse. In the framework, idle resources from vehicles, acting as CDC workers, are aggregated to handle intensive computation tasks in the vehicular metaverse. Specifically, in the upper layer, a miner coalition formation game is formulated based on a reputation metric to select reliable workers. To guarantee the reliable management of reputation values, the reputation values calculated based on the subjective logical model are maintained in a blockchain database. In the lower layer, a Stackelberg game based incentive mechanism is considered to attract reliable workers selected in the upper layer to participate in rendering tasks. The simulation results illustrate that the proposed framework is resistant to malicious workers. Compared with the baseline schemes, the proposed scheme can improve the utility of metaverse service provider and average profit of CDC workers.

TDTA: A truth detection based task assignment scheme for mobile crowdsourced Industrial Internet of Things

Mobile Crowdsensing (MCs), a promising strategy for intelligent data collection and task computing, is conducive to constructing industrial system and providing industrial services in the industrial Internet of Things (IIoT). How to ensure data reliability in an industrial environment is a challenging problem. In this article, we proposed a Truth Detection based Task Assignment (TDTA) scheme to assign micro-tasks to reliable workers and establish a credible task execution environment for crowdsourced IIoT. The TDTA scheme mainly contains three methods for different application scenarios: (1) Direct truth detection method. The Edge Node (EN) will calculate partial micro-tasks results as true results when idle, and compares them with the results reported by workers to detect their reliability directly. (2) Indirect truth detection method. After the platform obtains some trusted workers through the direct method, it distributes the same micro-tasks to workers with unknown credibility and trusted workers, and compares their results to verify reliability indirectly. (3) Post audit truth detection method. When idle, the platform recalculates historical reporting results for suspicious workers to verify their credibility. And the effect of time decay on credibility is considered. Moreover, the TDTA scheme also considers the factors of integrity and delay of the results to calculate the Quality of Service (QoS) value. Then the most active and credible workers are chosen to execute the task. Theoretical analysis and experiment results demonstrate the effectiveness of our proposed scheme, which has higher EN resource utilization, task result accuracy, and malicious worker recognition rate.

Secure Private and Adaptive Matrix Multiplication Beyond the Singleton Bound

We consider the problem of designing secure and private codes for distributed matrix-matrix multiplication. A master server owns two private matrices and hires worker nodes to help compute their product. The matrices should remain information-theoretically private from the workers. Some of the workers are malicious and return corrupted results to the master. We design a framework for security against malicious workers in private matrix-matrix multiplication. The main idea is a careful use of Freivalds’ algorithm to detect erroneous matrix multiplications. Our main goal is to apply this security framework to schemes with adaptive rates. Adaptive schemes divide the workers into clusters and thus provide flexibility in trading decoding complexity for efficiency. Our new scheme, SRPM3, provides a computationally efficient security check per cluster that detects the presence of one or more malicious workers with high probability. An additional per worker check is used to identify the malicious nodes. SRPM3 can tolerate the presence of an arbitrary number of malicious workers. We provide theoretical guarantees on the complexity of the security checks and simulation results on both, the missed detection rate as well as on the time needed for the integrity check.

FGFL: A blockchain-based fair incentive governor for Federated Learning

Federated Learning is a framework that coordinates a large amount of workers to train a shared model in a distributed manner, in which the training data are located on the workers' sides in order to preserve data privacy. There are two challenges in the crowdsourcing of FL, the workers who participant in training need to consume computing and communication resources, so that they are reluctant to participate in the training process if they can not get reasonable rewards. Moreover, there may be attackers who send arbitrary updates to get undeserving compensation or even destroy the model, thus, effective prevention of malicious workers is also critical. An incentive mechanism is urgently required in order to encourage high-quality workers to participate in FL and to punish the attackers. In this paper, we propose FGFL, a blockchain-based incentive governor for Federated Learning. In FGFL, we assess the participants with reputation and contribution indicators. Then the task publisher rewards workers fairly to attract efficient ones while the malicious ones are punished and eliminated. In addition, we propose a blockchain-based incentive management system to manage the incentive mechanism. We evaluate the effectiveness and fairness of FGFL through theoretical analysis and comprehensive experiments. The evaluation results show that FGFL fairly rewards workers according to their corresponding behavior and quality. FGFL increases the system revenue by 0.2% to 3.4% in reliable federations compared with baselines. And in the unreliable scenario where contains attackers, the system revenue of FGFL outperforms the baselines by more than 46.7%.

Stochastic alternating direction method of multipliers for Byzantine-robust distributed learning

This paper aims to solve a distributed learning problem under Byzantine attacks. In the underlying distributed master-worker architecture, there exist a number of unknown but malicious workers that can send arbitrary messages to the master to deviate the learning process, called Byzantine workers. In the literature, a total variation (TV) norm-penalized approximation formulation has been investigated to alleviate the effect of Byzantine attacks. To be specific, the TV norm penalty not only forces the local variables at the regular workers to be close, but is robust to the outliers sent by the Byzantine workers as well. For handling the separable TV norm-penalized approximation formulation, we propose a Byzantine-robust stochastic alternating direction method of multipliers (ADMM). Theoretically, we prove that the proposed method converges to a bounded neighborhood of the optimal solution at a rate of O(1/k) under mild assumptions, where k is the number of iterations and the size of neighborhood is determined by the number of Byzantine workers. Numerical experiments on the MNIST and COVERTYPE datasets further demonstrate the effectiveness of the proposed method to various Byzantine attacks.

A Lightweight and Privacy-Preserving Answer Collection Scheme for Mobile Crowdsourcing

Mobile Crowdsourcing (MCS) has become an emerging paradigm evolved from crowdsourcing by employing advanced features of mobile devices such as smartphones to perform more complicated, especially spatial tasks. One of the key procedures in MCS is to collect answers from mobile users (workers), which may face several security issues. First, authentication is required to ensure that answers are from authorized workers. In addition, MCS tasks are usually location-dependent, so the collected answers could disclose workers'' location privacy, which may discourage workers to participate in the tasks. Finally, the overhead occurred by authentication and privacy protection should be minimized since mobile devices are resource-constrained. Considering all the above concerns, in this paper, we propose a lightweight and privacy-preserving answer collection scheme for MCS. In the proposed scheme, we achieve anonymous authentication based on traceable ring signature, which provides authentication, anonymity, as well as traceability by enabling malicious workers tracing. In order to balance user location privacy and data availability, we propose a new concept named current location privacy, which means the location of the worker cannot be disclosed to anyone until a specified time. Since the leakage of current location will seriously threaten workers'' personal safety, causing such as absence or presence disclosure attacks, it is necessary to pay attention to the current location privacy of workers in MCS. We encrypt the collected answers based on timed-release encryption, ensuring the secure transmission and high availability of data, as well as preserving the current location privacy of workers. Finally, we analyze the security and performance of the proposed scheme. The experimental results show that the computation costs of a worker depend on the number of ring signature members, which indicates the flexibility for a worker to choose an appropriate size of the group under considerations of privacy and efficiency.

Remote Attestation on Behavioral Traces for Crowd Quality Control Based on Trusted Platform Module

Behavioral traces of workers have emerged as a new evidence to check the quality of their produced outputs in crowd computing. Whether the evidence is trustworthy or not is a key problem during the process. Challenges will be encountered in addressing this issue, because the evidence comes from unknown or adversarial workers. In this study, we proposed an alternative approach to ensure trustworthy evidence through a hardware-based remote attestation to bridge the gap. The integrity of the evidence was used as the trustworthy criterion. Trusted Platform Module (TPM) was considered the trusted anchor inspired by trusted computing to avoid unreliable or malicious workers. The module carefully recorded and stored many workers’ behavioral traces in the storage measurement log (SML). Each item in the log was extended to a platform configuration register (PCR) by the occurrence sequence of each event. The PCR was a tamper-proof storage inside the TPM. The value of the PCR was also considered evidence together with the SML. The evidence was sent to the crowdsourcing platform with the TPM signature. The platform checked the integrity of the evidence by a series of operations, such as validating the signature and recomputing the SML hash. This process was designed as a remote attestation protocol. The effectiveness, efficiency, and security of the protocol were verified theoretically and through experiments based on the open dataset, WebCrowd25K, and custom dataset. Results show that the proposed method is an alternative solution for ensuring the integrity of behavioral traces.

TruthTrust: Truth Inference-Based Trust Management Mechanism on a Crowdsourcing Platform.

On a crowdsourcing platform, in order to cheat for rewards or sabotage the crowdsourcing processes, spam workers may submit numerous erroneous answers to the tasks published by requesters. This type of behavior extremely reduces the completion rate of tasks and the enthusiasm of honest users, which may lead a crowdsourcing platform to a failure. Defending against malicious attacks is an important issue in crowdsourcing, which has been extensively addressed by existing methods, e.g., verification-based defense mechanisms, data analysis solutions, trust-based defense models, and workers’ properties matching mechanisms. However, verification-based defense mechanisms will consume a lot of resources, and data analysis solutions cannot motivate workers to provide high-quality services. Trust-based defense models and workers’ properties matching mechanisms cannot guarantee the authenticity of information when collusion requesters publish shadow tasks to help malicious workers get more participation opportunities. To defend such collusion attacks in crowdsourcing platforms, we propose a new defense model named TruthTrust. Firstly, we define a complete life cycle system that from users’ interaction to workers’ recommendation, and separately define the trust value of each worker and the credence of each requester. Secondly, in order to ensure the authenticity of the information, we establish a trust model based on the CRH framework. The calculated truth value and weight are used to define the global properties of workers and requesters. Moreover, we propose a reverse mechanism to improve the resistance under attacks. Finally, extensive experiments demonstrate that TruthTrust significantly outperforms the state-of-the-art approaches in terms of effective task completion rate.

A Novel Approach for Detection Insider Attacker Using Body Language

One of most important challenges in cyber security is detecting the insider attacker (a person or an employee with an authorized access to resources and data of an organization then uses that access - either wittingly or unwittingly - to harm organization security). This paper propose approach to obtain early indicator to insider attacker before doing the crime. The previous security systems focus on the technical anomaly of an employee to discover the insider attacker and can’t discover it, if there isn’t technical anomaly. This paper attempt to discover insider attacker when there isn’t technical anomaly, Where presented body language-based approach to give earlier alarm of insider attacker. By using three of negative body language gestures (Cross Arms, Clasped Hands, Covering the Mouth) which referred to feeling of insecure, ready for an attack, doubt and a lack of self-confidence, these feelings are the closest to the feelings of the internal attacker. These gestures obtained by use skeleton features from video stream provided by Orbbec Astra Pro camera after passed to rule based classifier to recognize each one of the three body language gestures. Then determined the degree of trust based on the duration of the gesture and the number of occurrences of the same gesture or different gestures and depending on the degree of trust, the organization is alerted to the questionable employees. The test performs on ten of employees, four insider attackers were planted among them, and the results show 70% accuracy of detects the insiders, this approach will detect insider attacker before started his malicious work.

A Novel Hadoop Security Model for Addressing Malicious Collusive Workers.

With the daily increase of data production and collection, Hadoop is a platform for processing big data on a distributed system. A master node globally manages running jobs, whereas worker nodes process partitions of the data locally. Hadoop uses MapReduce as an effective computing model. However, Hadoop experiences a high level of security vulnerability over hybrid and public clouds. Specially, several workers can fake results without actually processing their portions of the data. Several redundancy-based approaches have been proposed to counteract this risk. A replication mechanism is used to duplicate all or some of the tasks over multiple workers (nodes). A drawback of such approaches is that they generate a high overhead over the cluster. Additionally, malicious workers can behave well for a long period of time and attack later. This paper presents a novel model to enhance the security of the cloud environment against untrusted workers. A new component called malicious workers' trap (MWT) is developed to run on the master node to detect malicious (noncollusive and collusive) workers as they convert and attack the system. An implementation to test the proposed model and to analyze the performance of the system shows that the proposed model can accurately detect malicious workers with minor processing overhead compared to vanilla MapReduce and Verifiable MapReduce (V-MR) model [1]. In addition, MWT maintains a balance between the security and usability of the Hadoop cluster.

ATM: Attribute-Based Privacy-Preserving Task Assignment and Incentive Mechanism for Crowdsensing

Crowdsensing is a practical component in Internet-of-Things, in which task requesters outsource sensing tasks to workers by a crowdsensing server (CS). Task assignment and incentive design are two essential parts of crowdsensing. However, due to the semi-honest CS and malicious workers, there exist several security and fairness issues. First, during task assigning and reward distributing, CS can get the private information of both task requester and workers such as identity, attributes. Second, malicious workers who are unable to execute tasks can get the task content, which leaks task privacy. Third, most auction-based incentive mechanism only considers the biddings of workers and ignores the real capabilities of workers. To this end, we propose a privacy-preserving fine-grained task assignment scheme for the crowdsensing system. Moreover, we also design a novel incentive mechanism based on the capabilities of workers. To be specific, a ciphertext-policy attribute-based encryption (CP-ABE) scheme with the characteristics of hidden policy is adopted to select workers and protect the privacy that of both requirements and workers’ attributes. We quantify the capabilities of workers based on the workers’ attributes and further determine the workers’ reward according to their capabilities. Detailed security analysis demonstrates that our proposed scheme can protect workers’ identity privacy, attribute privacy, and data confidentiality. Besides, the extensive experiments show that our incentive mechanism can effectively motivate workers.

On the Initial Behavior Monitoring Issues in Federated Learning

In Federated Learning (FL), a group of workers participate to build a global model under the coordination of one node, the chief. Regarding the cybersecurity of FL, some attacks aim at injecting the fabricated local model updates into the system. Some defenses are based on malicious worker detection and behavioral pattern analysis. In this context, without timely and dynamic monitoring methods, the chief cannot detect and remove the malicious or unreliable workers from the system. Our work emphasize the urgency to prepare the federated learning process for monitoring and eventually behavioral pattern analysis. We study the information inside the learning process in the early stages of training, propose a monitoring process and evaluate the monitoring period required. The aim is to analyse at what time is it appropriate to start the detection algorithm in order to remove the malicious or unreliable workers from the system and optimise the defense mechanism deployment. We tested our strategy on a behavioral pattern analysis defense applied to the FL process of different benchmark systems for text and image classification. Our results show that the monitoring process lowers false positives and false negatives and consequently increases system efficiency by enabling the distributed learning system to achieve better performance in the early stage of training.

Context-Aware Reliable Crowdsourcing in Social Networks

There are two problems in the traditional crowdsourcing systems for handling complex tasks. First, decomposing complex tasks into a set of micro-subtasks requires the decomposition capability of the requesters; thus, some requesters may abandon using crowdsourcing to accomplish a large number of complex tasks since they cannot bear such heavy burden by themselves. Second, tasks are often assigned redundantly to multiple workers to achieve reliable results, but reliability may not be ensured when there are many malicious workers in the crowd. Currently, it is observed that the workers are often connected through social networks, a feature that can significantly facilitate task allocation and task execution in crowdsourcing. Therefore, this paper investigates crowdsourcing in social networks and presents a novel context-aware reliable crowdsourcing approach. In our presented approach, the two problems in traditional crowdsourcing are addressed as follows: 1) the complex tasks can be performed through autonomous coordination between the assigned worker and his contextual workers in the social network; thus, the requesters can be exempt from a heavy computing load for decomposing complex tasks into subtasks and combing the partial results of subtasks, thereby enabling more requesters to accomplish a large number of complex tasks through crowdsourcing, and 2) the reliability of a worker is determined not only by the reputation of the worker himself but also by the reputations of the contextual workers in the social network; thus, the unreliability of transient or malicious workers can be effectively addressed. The presented approach addresses two types of social networks including simplex and multiplex networks. Based on theoretical analyses and experiments on a real-world dataset, we find that the presented approach can achieve significantly higher task allocation and execution efficiency than the previous benchmark task allocation approaches; moreover, the presented contextual reputation mechanism can achieve relatively higher reliability when there are many malicious workers in the crowd.

Deep Reinforcement Learning for Partially Observable Data Poisoning Attack in Crowdsensing Systems

Crowdsensing systems collect various types of data from sensors embedded on mobile devices owned by individuals. These individuals are commonly referred to as workers that complete tasks published by crowdsensing systems. Because of the relative lack of control over worker identities, crowdsensing systems are susceptible to data poisoning attacks which interfering with data analysis results by injecting fake data conflicting with ground truth. Frameworks like TruthFinder can resolve data conflicts by evaluating the trustworthiness of the data providers. These frameworks somehow make crowdsensing systems more robust since they can limit the impact of dirty data by reducing the value of unreliable workers. However, previous work has shown that TruthFinder may also be affected by the data poisoning attack when the malicious workers have access to global information. In this article, we focus on partially observable data poisoning attacks in crowdsensing systems. We show that even if the malicious workers only have access to local information, they can find effective data poisoning attack strategies to interfere with crowdsensing systems with TruthFinder. First, we formally model the problem of partially observable data poisoning attack against crowdsensing systems. Then, we propose a data poisoning attack method based on deep reinforcement learning, which helps malicious workers jeopardize with TruthFinder while hiding themselves. Based on the method, the malicious workers can learn from their attack attempts and evolve the poisoning strategies continuously. Finally, we conduct experiments on real-life data sets to verify the effectiveness of the proposed method.