Federated learning is a popular framework designed to perform the distributed machine while protecting client privacy. However, heterogeneous data distribution in real-world environments makes it difficult converge when performing model training. In this article, we propose federated gradient scheduling (FedGS), an improved historical sampling utilization method for optimizers that utilize gradients alleviate instability problem of information due non-IID. FedGS consists two main steps...

Centralized learning now faces data mapping and security constraints that make it difficult to carry out. Federated with a distributed architecture has changed this situation. By restricting the training process participants' local, federated addresses model needs of multiple sources while better protecting privacy. However, in real-world application scenarios, need achieve fairness addition privacy protection. In practice, could happen some participants specific motives may short join...

Federated learning (FL) provides a framework without participants sharing local raw data, but individual privacy is still at risk of disclosure through attacking the trained models. Due to strong guarantee, differential (DP) widely applied FL avoid leakage. Traditional private adds noise directly gradients. The continuous accumulated on parameter models severely impairs effectiveness. To solve this problem, we introduce idea differentially data release (DPCR) into and propose an based DPCR...

This paper proposes a maze exploring algorithm named "Partition-central Algorithm", which is used to find the shortest path in micromouse competition maze. A standard 16*16 units divided into 12 partitions this algorithm. Depending on absolute direction of and locations each partition, rules alter when walks optimize process. simulation program developed verify The test result shows that Partition-central has higher average efficiency compared with other algorithms.

Global synchronization in adaptive coupling networks is studied this paper. A new simple controller proposed based on a concept of asymptotically stable led by partial state variables. Under the update law, network can achieve global without calculating eigenvalues outer matrix. The law only dependent variables individual oscillators. Numerical simulations are given to show effectiveness method, which unified chaotic system chosen as nodes with different topologies.

Federated learning (FL) is a novel machine framework in which models are built jointly by multiple parties. We investigate the privacy preservation of XGBoost, gradient boosting decision tree (GBDT) model, context FL. While recent work relies on cryptographic schemes to preserve model gradients, these methods computationally expensive. In this paper, we propose an adaptive privacy-preserving algorithm based differential (DP), more efficient. Our perturbs individual data computing mean per...

Integrating data from multiple parties to achieve cross-institutional machine learning is an important trend in Industry 4.0 era. However, the privacy risks sharing pose a significant challenge integration. To integrate without and meet large-scale samples’ modeling needs, we propose two vertical federation algorithms for ridge regression via least-squares solution two-party multi-party scenarios, respectively. Compared with state-of-the-art algorithms, our only need one round of calculation...

Currently, financial institutions utilize personal sensitive information extensively in machine learning. It results significant privacy risks to customers. As an essential standard of privacy, differential is often applied learning recent years. To establish a prediction model credit card default under the premise protecting we consider problems customer data contribution difference and sample distribution imbalance, propose weighted SVM algorithm privacy. Through theoretical analysis, have...

Software license is the most well-used technology in copyright protection of commercial software and shareware, as well development management free open source software. There are several different types that allow controlled access services. The two popular fixed license, which gives rights for an identified workstation, floating restricts number simultaneous users to a certain bound. In this paper, we propose system, called MFFLS, efficiently implements

Hierarchical trees are widely used in differentially private statistical releases. Existing fast specialized algorithms guarantee hierarchical consistency but not non-negativity, which may incur meaningless negative values due to randomness. Quadratic programming can achieve optimally non-negative consistent releases, traditional numerical-based methods face significant performance challenges when handling large-scale trees. In this article, we construct the element set of Lagrange...

Object detection has achieved significant progress in attaining high-quality performance without leaking private messages. However, traditional approaches cannot defend the poisoning attacks. Poisoning attacks can make predictive model unusable, which quickly causes recognition errors or even traffic accidents. In this paper, we propose a privacy-preserving object with (PR-PPOD) framework via distributed training help of CNN, ResNet18, and classical SSD network. Specifically, design...

Being the core course of computer science, teaching "the Principle Computer Organization", concerning its training purpose and objective, should focus on nurturing students' ability discovering, analyzing solving problems with an overall perspective course, so as to lay a solid foundation for systematic wholeness improve professional well. And concept course-group is also proposed in paper, which means joint construction correlated courses under frame science syllabus unified planning,...

This paper introduces several new guiding ideologies in "the Principle of Computer Organization" course reform, based on the purpose application-oriented personnel training, namely teaching according to students aptitude, focusing students' capacity-building, emphasizing both theories and experiments. In practice curriculum teaching, we focus training enhancing self-learning ability through group collaborative learning examination reform. Meanwhile, try motivate study with process management...

When using differential privacy to publish high‐dimensional data, the huge dimensionality leads greater noise. Especially for binary it is easy be covered by excessive Most existing methods cannot address real data problems appropriately because they suffer from high time complexity. Therefore, in response above, we propose adaptive Bayesian network algorithm PrivABN data. This uses a new greedy accelerate construction of networks, which reduces complexity GreedyBayes O ( n m 4 ). In...