Search

Domain Generation Algorithms (DGAs) are used by malware to generate pseudorandom domain names to establish communication between infected bots and command and control servers. While DGAs can be detected by machine learning (ML) models with great accuracy, offering DGA detection as a service raises privacy concerns when requiring network administrators to disclose their DNS traffic to the service provider. The main scientific contribution of this paper is to propose the first end-to-end framework for privacy-preserving classification as a service of domain names into DGA (malicious) or non-DGA (benign) domains. Our framework achieves these goals by carefully designed protocols that combine two privacy-enhancing technologies (PETs), namely secure multi-party computation (MPC) and differential privacy (DP). Through MPC, our framework enables an enterprise network administrator to outsource the problem of classifying a DNS (Domain Name System) domain as DGA or non-DGA to an external organization without revealing any information about the domain name. Moreover, the service provider's ML model used for DGA detection is never revealed to the network administrator. Furthermore, by using DP, we also ensure that the classification result cannot be used to learn information about individual entries of the training data. Finally, we leverage post-training float16 quantization of deep learning models in MPC to achieve efficient, secure DGA detection. We demonstrate that by using quantization achieves a significant speed-up, resulting in a 23% to 42% reduction in inference runtime without reducing accuracy using a three party secure computation protocol tolerating one corruption. Previous solutions are not end-to-end private, do not provide differential privacy guarantees for the model's outputs, and assume that model embeddings are publicly known. Our best protocol in terms of accuracy runs in about 0.22s. [ABSTRACT FROM AUTHOR]

Background: Cancer diagnosis based on machine learning has become a popular application direction. Support vector machine (SVM), as a classical machine learning algorithm, has been widely used in cancer diagnosis because of its advantages in high-dimensional and small sample data. However, due to the high-dimensional feature space and high feature redundancy of gene expression data, SVM faces the problem of poor classification effect when dealing with such data. Methods: Based on this, this paper proposes a hybrid feature selection algorithm combining information gain and grouping particle swarm optimization (IG-GPSO). The algorithm firstly calculates the information gain values of the features and ranks them in descending order according to the value. Then, ranked features are grouped according to the information index, so that the features in the group are close, and the features outside the group are sparse. Finally, grouped features are searched using grouping PSO and evaluated according to in-group and out-group. Results: Experimental results show that the average accuracy (ACC) of the SVM on the feature subset selected by the IG-GPSO is 98.50%, which is significantly better than the traditional feature selection algorithm. Compared with KNN, the classification effect of the feature subset selected by the IG-GPSO is still optimal. In addition, the results of multiple comparison tests show that the feature selection effect of the IG-GPSO is significantly better than that of traditional feature selection algorithms. Conclusion: The feature subset selected by IG-GPSO not only has the best classification effect, but also has the least feature scale (FS). More importantly, the IG-GPSO significantly improves the ACC of SVM in cancer diagnostic. [ABSTRACT FROM AUTHOR]

This paper presents a multi-algorithm fusion model (StackingGroup) based on the Stacking ensemble learning framework to address the variable selection problem in high-dimensional group structure data. The proposed algorithm takes into account the differences in data observation and training principles of different algorithms. It leverages the strengths of each model and incorporates Stacking ensemble learning with multiple group structure regularization methods. The main approach involves dividing the data set into K parts on average, using more than 10 algorithms as basic learning models, and selecting the base learner based on low correlation, strong prediction ability, and small model error. Finally, we selected the grSubset + grLasso, grLasso, and grSCAD algorithms as the base learners for the Stacking algorithm. The Lasso algorithm was used as the meta-learner to create a comprehensive algorithm called StackingGroup. This algorithm is designed to handle high-dimensional group structure data. Simulation experiments showed that the proposed method outperformed other R2, RMSE, and MAE prediction methods. Lastly, we applied the proposed algorithm to investigate the risk factors of low birth weight in infants and young children. The final results demonstrate that the proposed method achieves a mean absolute error (MAE) of 0.508 and a root mean square error (RMSE) of 0.668. The obtained values are smaller compared to those obtained from a single model, indicating that the proposed method surpasses other algorithms in terms of prediction accuracy. [ABSTRACT FROM AUTHOR]

The paper presents an algorithm to approach the problem of Maximum Clique Enumeration, a well known NP-hard problem that have several real world applications. The proposed solution, called LGP-MCE, exploits Geometric Deep Learning, a Machine Learning technique on graphs, to filter out nodes that do not belong to maximum cliques and then applies an exact algorithm to the pruned network. To assess the LGP-MCE, we conducted multiple experiments using a substantial dataset of real-world networks, varying in size, density, and other characteristics. We show that LGP-MCE is able to drastically reduce the running time, while retaining all the maximum cliques. [ABSTRACT FROM AUTHOR]

Recommendation algorithms save a lot of valuable time for people to get the information they are interested in. However, the feature calculation and extraction process of each machine learning or deep learning recommendation algorithm are different, so how to obtain various features with different dimensions, i.e., how to integrate the advantages of each model and improve the model inference efficiency, becomes the focus of this paper. In this paper, a better deep learning model is obtained by integrating several cutting-edge deep learning models. Meanwhile, to make the integrated learning model converge better and faster, the parameters of the integrated module are initialized, constraints are imposed, and a new activation function is designed for better integration of the sub-models. Finally, the integrated large model is distilled for knowledge distillation, which greatly reduces the number of model parameters and improves the model inference efficiency. [ABSTRACT FROM AUTHOR]

Q-learning is a classical reinforcement learning algorithm and one of the most important methods of mobile robot path planning without a prior environmental model. Nevertheless, Q-learning is too simple when initializing Q-table and wastes too much time in the exploration process, causing a slow convergence speed. This paper proposes a new Q-learning algorithm called the Paired Whale Optimization Q-learning Algorithm (PWOQLA) which includes four improvements. Firstly, to accelerate the convergence speed of Q-learning, a whale optimization algorithm is used to initialize the values of a Q-table. Before the exploration process, a Q-table which contains previous experience is learned to improve algorithm efficiency. Secondly, to improve the local exploitation capability of the whale optimization algorithm, a paired whale optimization algorithm is proposed in combination with a pairing strategy to speed up the search for prey. Thirdly, to improve the exploration efficiency of Q-learning and reduce the number of useless explorations, a new selective exploration strategy is introduced which considers the relationship between current position and target position. Fourthly, in order to balance the exploration and exploitation capabilities of Q-learning so that it focuses on exploration in the early stage and on exploitation in the later stage, a nonlinear function is designed which changes the value of ε in ε-greedy Q-learning dynamically based on the number of iterations. Comparing the performance of PWOQLA with other path planning algorithms, experimental results demonstrate that PWOQLA achieves a higher level of accuracy and a faster convergence speed than existing counterparts in mobile robot path planning. The code will be released at https://github.com/wanghanyu0526/improveQL.git. [ABSTRACT FROM AUTHOR]

Kmeans clustering algorithm is an iterative unsupervised learning algorithm that tries to partition the given dataset into k pre-defined distinct non-overlapping clusters where each data point belongs to only one group. However, its performance is affected by its sensitivity to the initial cluster centroids with the possibility of convergence into local optimum and specification of cluster number as the input parameter. Recently, the hybridization of metaheuristics algorithms with the K-Means algorithm has been explored to address these problems and effectively improve the algorithm's performance. Nonetheless, most metaheuristics algorithms require rigorous parameter tunning to achieve an optimum result. This paper proposes a hybrid clustering method that combines the well-known symbiotic organisms search algorithm with K-Means using the SOS as a global search metaheuristic for generating the optimum initial cluster centroids for the K-Means. The SOS algorithm is more of a parameter-free metaheuristic with excellent search quality that only requires initialising a single control parameter. The performance of the proposed algorithm is investigated by comparing it with the classical SOS, classical K-means and other existing hybrids clustering algorithms on eleven (11) UCI Machine Learning Repository datasets and one artificial dataset. The results from the extensive computational experimentation show improved performance of the hybrid SOSK-Means for solving automatic clustering compared to the standard K-Means, symbiotic organisms search clustering methods and other hybrid clustering approaches. [ABSTRACT FROM AUTHOR]

Every year, millions of new devices are added to the Internet of things, which has both great benefits and serious security risks for user data privacy. It is the device owners' responsibility to ensure that the ownership settings of Internet of things devices are maintained, allowing them to communicate with other user devices autonomously. The ultimate goal of the future Internet of Things is for it to be able to make decisions on its own, without the need for human intervention. Therefore, trust computing and prediction have become more vital in the processing and handling of data as well as in the delivery of services. In this paper, we compute trust in social IoT scenarios using a hybrid approach that combines a distributed computation technique and a global machine learning approach. The approach considers social similarity while assessing other users' ratings and utilize a cloud-based architecture. Further, we propose a dynamic way to aggregate the different computed trust values. According to the results of the experimental work, it is shown that the proposed approaches outperform related work. Besides, it is shown that the use of machine learning provides slightly better performance than the computing model. Both proposed approaches were found successful in degrading malicious ratings without the need for more complex algorithms. [ABSTRACT FROM AUTHOR]

This paper focuses on 6D pose estimation for weakly textured targets from RGB-D images. A 6D pose estimation algorithm (DOPE++) based on a deep neural network for weakly textured objects is proposed to solve the poor real-time pose estimation and low recognition efficiency in the robot grasping process of parts with weak texture. More specifically, we first introduce the depthwise separable convolution operation to lighten the original deep object pose estimation (DOPE) network structure to improve the network operation speed. Second, an attention mechanism is introduced to improve network accuracy. In response to the low recognition efficiency of the original DOPE network for parts with occlusion relationships and the false recognition problem in recognizing parts with scales that are too large or too small, a random mask local processing method and a multiscale fusion pose estimation module are proposed. The results show that our proposed DOPE++ network improves the real-time performance of 6D pose estimation and enhances the recognition of parts at different scales without loss of accuracy. To address the problem of a single background representation of the part pose estimation dataset, a virtual dataset is constructed for data expansion to form a hybrid dataset. [ABSTRACT FROM AUTHOR]

Measurements of the physical outputs of speech—vocal tract geometry and acoustic energy—are high-dimensional, but linguistic theories posit a low-dimensional set of categories such as phonemes and phrase types. How can it be determined when and where in high-dimensional articulatory and acoustic signals there is information related to theoretical categories? For a variety of reasons, it is problematic to directly quantify mutual information between hypothesized categories and signals. To address this issue, a multi-scale analysis method is proposed for localizing category-related information in an ensemble of speech signals using machine learning algorithms. By analyzing how classification accuracy on unseen data varies as the temporal extent of training input is systematically restricted, inferences can be drawn regarding the temporal distribution of category-related information. The method can also be used to investigate redundancy between subsets of signal dimensions. Two types of theoretical categories are examined in this paper: phonemic/gestural categories and syntactic relative clause categories. Moreover, two different machine learning algorithms were examined: linear discriminant analysis and neural networks with long short-term memory units. Both algorithms detected category-related information earlier and later in signals than would be expected given standard theoretical assumptions about when linguistic categories should influence speech. The neural network algorithm was able to identify category-related information to a greater extent than the discriminant analyses. [ABSTRACT FROM AUTHOR]