As a decentralized machine learning approach, federated learning has become a key solution for numerous applications. However, one primary challenge lies in the data heterogeneity issue when developing effective feder...
As a decentralized machine learning approach, federated learning has become a key solution for numerous applications. However, one primary challenge lies in the data heterogeneity issue when developing effective federated learning algorithms. While many existing frameworks address the concern of non-independent and identically distributed data(non-IID), they have not considered heterogeneity of data quality. Therefore, we propose ROCFL, a robust clustered federated learning method in this study, which can amplify the disparity in weight allocation between models trained on different quality data. We first develop an optimal clustering matching mechanism that groups clients with similar data distributions. This allows the optimal clustering model to be derived without specifying a predetermined number of clusters. Next, we introduce a personalized weight allocation strategy, which assigns a weight benchmark to each cluster based on its cluster importance index. This strategy mitigates the negative impacts of low-quality data during model aggregation. Finally, we design a federated aggregation strategy grounded in a sampling approach. It not only ensures unbiased sampling but also significantly reduces both computational and communication overheads. Experiments have been carried out and the results demonstrate that higher accuracy and robustness can be achieved by ROCFL in scenarios wherein data distribution and quality heterogeneity coexist, in comparison with the benchmarks.
Classical novae are cataclysmic binary star systems in which the matter of a companion star is accreted on a white dwarf1,2. Accumulation of hydrogen in a layer eventually causes a thermonuclear explosion on the surfa...
Classical novae are cataclysmic binary star systems in which the matter of a companion star is accreted on a white dwarf1,2. Accumulation of hydrogen in a layer eventually causes a thermonuclear explosion on the surface of the white dwarf3, brightening the white dwarf to ~105 solar luminosities and triggering ejection of the accumulated matter. Novae provide the extreme conditions required to accelerate particles, electrons or protons, to high energies. Here we present the detection of gamma rays by the MAGIC telescopes from the 2021 outburst of RS Ophiuchi, a recurrent nova with a red giant companion, which allowed us to accurately characterize the emission from a nova in the 60 GeV to 250 GeV energy range. The theoretical interpretation of the combined Fermi LAT and MAGIC data suggests that protons are accelerated to hundreds of gigaelectronvolts in the nova shock. Such protons should create bubbles of enhanced cosmic ray density, of the order of 10 pc, from the recurrent *** of the 2021 outburst of the nova RS Oph in very-high-energy gamma rays by the MAGIC telescopes is reported. Investigation of the gamma-ray emission provides evidence for acceleration of protons within the nova shock, which then propagate outwards to create bubbles of enhanced cosmic ray density.
The time-space transferability of data center workload determines that its power load has considerable adjustable potential. Making full use of this potential to formulate power load transfer strategy and participatin...
详细信息
The time-space transferability of data center workload determines that its power load has considerable adjustable potential. Making full use of this potential to formulate power load transfer strategy and participating in demand response or electricity market can effectively reduce the operating cost of data center. Accurate load forecasting results of data center are important supports for formulating effective transfer strategy. However, existing data center load forecasting methods still do not fully consider the influencing factors of different types of loads in the data center. And they are inapplicable when behind-the-meter (BTM) distributed photovoltaics (DPVs) are installed in the data center. To solve this, this paper divides the data center net load into two parts and forecasting them respectively: the cooling load (CL) and the remaining net load (RNL), where RNL is the sum of IT load, DPV output and other power loads. According to the characteristics of CL and DPV, we divided CL and RNL into seasons and periods respectively. Under each season or period, feature engineering based on light gradient boosting machine (LightGBM) is created and the LightGBM is also used to establish the forecasting model. The selection of parameters is realized by grid search algorithm (GS). Finally, the validity of the method is verified by the load data provided by National Renewable Energy laboratory (NREL).
In this paper, we study the band structure engineering of black phosphorus/graphene/MoS2 (BP/graphene/MoS2) van der Waals heterojunctions based on ab initio simulations. The density of state and charge density of BP/M...
详细信息
The similarity of binary code is widely used in code copyright protection, vulnerability mining, malicious code analysis and etc. In this paper, we proposed a method for measuring/evaluating the similarity of two bina...
详细信息
To many people, smartphones and other mobile devices have been an indispensable part of their daily lives. With personal financial and medical information and job-related data processed on mobile devices,systems and n...
To many people, smartphones and other mobile devices have been an indispensable part of their daily lives. With personal financial and medical information and job-related data processed on mobile devices,systems and networks as well as the connected mobile cloud, it is critically important that secure mobile operating systems, secure mobile applications, secure smartphone devices, secure cellular networks and secure mobile cloud all keep up with the ever-growing mobile usage.
The rapid increase in Android malware has brought great challenges to malware analysis. To deal with such a severe situation, it has been proposed an effective way which groups malware with common behaviors into the s...
详细信息
ISBN:
(数字)9781728195537
ISBN:
(纸本)9781728195544
The rapid increase in Android malware has brought great challenges to malware analysis. To deal with such a severe situation, it has been proposed an effective way which groups malware with common behaviors into the same malware family. Although there are many methods for malware family classification, the most critical and primary step is always the definition of sensitive behavior in an application, which will be beneficial for the later classification task. Much existing literature has manually selected sensitive features, such as permission, or even designed graph-based features via the control flow graph. They heavily depend on expert knowledge and time-consuming malware application analysis, which means it has to focus on the mal ware itself to dig out valuable security knowledge at first. However, the zooming malware overwhelms such expensive feature definition methods. To overcome such a problem, we adopt a knowledge graph-based sensitive feature selection method for Android mal ware classification. Based on the Android Developer documentation, an Android API knowledge graph is constructed at first. We can obtain not only permission but also related critical API from this graph. Note that both hyperlink relation and similarity relation are used to find out the critical API. With the knowledge graph-based sensitive features, we represent each Android malware as a boolean feature vector and send it in to a machine learning classifier for malware classification. We evaluate our proposed methods on three well-known Android malware datasets, such as Genome, Drebin, and AMD. The experimental results show that: 1) our proposed sensitive API is advantageous for malware detection; 2) API chosen by similarity relation can marginally improve performance; 3) different permission groups also make an influence for classification.
The similarity of binary code is widely used in code copyright protection, vulnerability mining, malicious code analysis and etc. In this paper, we proposed a method for measuring/evaluating the similarity of two bina...
详细信息
The similarity of binary code is widely used in code copyright protection, vulnerability mining, malicious code analysis and etc. In this paper, we proposed a method for measuring/evaluating the similarity of two binary files based on software genes. Some of Natural language processing methods were introduced into program semantics analysis, including word2vec and doc2vec models to generate assembly instruction embeddings and gene semantic embeddings. Then the longest common subsequence method was applied to evaluate the software similarity. Experiments show that our method can effectively evaluate the similarity of binary files.
暂无评论