Kind Of Data Distribution Research Articles

FLEE: A Hierarchical Federated Learning Framework for Distributed Deep Neural Network over Cloud, Edge, and End Device

With the development of smart devices, the computing capabilities of portable end devices such as mobile phones have been greatly enhanced. Meanwhile, traditional cloud computing faces great challenges caused by privacy-leakage and time-delay problems, there is a trend to push models down to edges and end devices. However, due to the limitation of computing resource, it is difficult for end devices to complete complex computing tasks alone. Therefore, this article divides the model into two parts and deploys them on multiple end devices and edges, respectively. Meanwhile, an early exit is set to reduce computing resource overhead, forming a hierarchical distributed architecture. In order to enable the distributed model to continuously evolve by using new data generated by end devices, we comprehensively consider various data distributions on end devices and edges, proposing a hierarchical federated learning framework FLEE , which can realize dynamical updates of models without redeploying them. Through image and sentence classification experiments, we verify that it can improve model performances under all kinds of data distributions, and prove that compared with other frameworks, the models trained by FLEE consume less global computing resource in the inference stage.

Instability results for Euclidean distance, nearest neighbor search on high dimensional Gaussian data

In 1998, Beyer et al. described a nearest neighbor query as unstable if the query point has nearly identical distance from all points in the dataset. Subsequently, researchers have proven that, as data dimensionality goes to infinity, the probability of query instability approaches one for various kinds of data distributions, dataset size functions, and distance metrics. This paper addresses the problem of characterizing query instability behavior over centered Gaussian data generation distributions and Euclidean distance. Sufficient conditions are established on the covariance matrices and dataset size function under which the probability of query instability approaches one. Furthermore, conditions are also established under which the query instability probability is strictly bounded away from one for a non-vanishing set of query points.

A survey on effects of class imbalance in data pre-processing stage of classification problem

Classifier learning with datasets suffering from imbalance class distribution is a challenging task and it hinders the performance of machine learning algorithms. This imbalance occurs when a particular class is highly outnumbered than that of another class. Such kind of data distribution in the real world applications caught the attention of many researchers. This paper presents the review of various state of the art sampling techniques and ensemble techniques to resolve class imbalance. This paper also investigates the other factors such as threshold of distribution, inter or within class imbalance, etc., that make class imbalance a more complex issue. Comparisons of various approaches viz. data sampling, cost sensitive methods, bagging, boosting which alleviate the class imbalance problem are investigated in detail for their effects on class imbalance problem. Different parameters have been reviewed for measuring and evaluating the performance of the model. Accuracy is majorly used as evaluation parameter in machine learning problems, but from reviews it is found that there are different parameters such as precision, recall and AU-ROC which provide statistical measures for evaluating the model. The paper gives research directions in the domain of class imbalance problems.

Comparative Investigations and Performance Evaluation for Multiple-Level Association Rules Mining Algorithm

This paper focuses on the comparative investigation and performance evaluation of the ML_TMLA algorithm that generates multiple transaction tables for all levels in one database scan with that of ML_T2L1 and ML_T1LA algorithms. The performance study has been carried out on different kinds of data distributions (three synthetic and one real dataset) and thresholds that identify the conditions for algorithm selection. The AR Tool has been used for the experimental and comparative evaluation of the proposed algorithm with other algorithms.

Towards data mining benchmarking

Performance benchmarking has played an important role in the research and development in relational DBMS, object-relational DBMS, data warehouse systems, etc. We believe that benchmarking data mining algorithms is a long overdue task, and it will play an important role in the research and development of data mining systems as well. Frequent pattern mining forms a core component in mining associations, correlations, sequential patterns, partial periodicity, etc., which are of great potential value in applications. There have been a lot of methods proposed and developed for efficient frequent pattern mining in various kinds of databases, including transaction databases, time-series databases, etc. However, so far there is no serious performance benchmarking study of different frequent pattern mining methods. To facilitate an analytical comparison of different frequent mining methods, we have constructed an open test bed for performance study of a set of recently developed, popularly used methods for mining frequent patterns in transaction databases and mining sequential patterns in sequence databases, with different data characteristics. The testbed consists of the following components. A synthetic data generator, which can generate large sets of synthetic data in various kinds of data distributions. A few large data sets from real world applications will also be provided. A good set of typical frequent pattern mining methods, ranging from classical algorithms to recent studies. The method are grouped into three classes: frequent pattern mining, max-pattern mining , and sequential pattern mining . For frequent pattern mining, we will demonstrate Apriori, hashing, partitioning, sampling, TreeProjection, and FP-growth. For maximal pattern mining, we will demonstrate MaxMiner, TreeProjection, and FP-growth-max. For sequential pattern mining, we will demonstrate GSP and FreeSpan. A set of performance curves. These algorithms their running speeds, scalabilities, bottlenecks, and performance on different data distributions, will be compared and demonstrated upon request. Some performance curves from our pre-conference experimental evaluations will also be shown. An open testbed. Our goal is to construct an extensible test bed which integrates the above components and supports an open-ended testing service. Researchers can upload the object codes of their mining algorithms, and run them in the test bed using these data sets. The architecture is shown in Figure 1. This testbed is our first step towards benchmarking data mining algorithms. By doing so, performance of different algorithms can be reported consistently, on the same platform, and in the same environment. After the demo, we plan to make the testbed available on the WWW so that it may, hopefully, benefit further research and development of efficient data mining methods.