Grid‐based high performance ensemble classification for evolving data stream

Qi, Qian; Xie, Mengbo; Xiao, Chao-Jie; Zhang, Rui

doi:10.1002/cpe.3898

Cited by 5 publications

(8 citation statements)

References 15 publications

(24 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Among the works that explored multi-core parallelism, distributed or not, we can further subdivide it into batch [9,18,21,22,25,44] or data stream [20,30,31,37] methods. Many works with various ensemble methods used the Message Passing Interface (MPI) standard, such as for ensembles of improved and faster Support Vector Machine (SVM) [18], bagging decision rule ensembles [20] and regression ensembles [31].…”

Section: Related Workmentioning

confidence: 99%

“…In [25], an efficient Random Forest (RF) implementation that improves memory access due to better data representation on machines that combine both shared and distributed memory is proposed and implemented using FREERIDE (previous work from the authors). In [37], an ensemble of J48 is parallelized for grid platforms using Java. In [30], a low-latency Hoeffding Tree (HT) is implemented in C++ and used in RFs.…”

Section: Related Workmentioning

confidence: 99%

“…Qian et al [37] propose a novel ensemble for data stream classification. This solution maps different raw data to multiple grids, where the first-order geometric center is used to represent and classify data.…”

Section: Related Workmentioning

confidence: 99%

“…The works in [20,30] focus on specific algorithms, SCALLOP and Binary HT, respectively. The work in [31] leverages parallel processing to improve the algorithm's parameters, while [37] is focused on data compression.…”

Section: Related Workmentioning

confidence: 99%

See 3 more Smart Citations

Improving the performance of bagging ensembles for data streams through mini-batching

Cassales,

Gomes,

Bifet

et al. 2021

Preprint

View full text Add to dashboard Cite

Often, machine learning applications have to cope with dynamic environments where data are collected in the form of continuous data streams with potentially infinite length and transient behavior. Compared to traditional (batch) data mining, stream processing algorithms have additional requirements regarding computational resources and adaptability to data evolution. They must process instances incrementally because the data's continuous flow prohibits storing data for multiple passes. Ensemble learning achieved remarkable predictive performance in this scenario. Implemented as a set of (several) individual classifiers, ensembles are naturally amendable for task parallelism. However, the incremental learning and dynamic data structures used to capture the concept drift increase the cache misses and hinder the benefit of parallelism. This paper proposes a mini-batching strategy that can improve memory access locality and performance of several ensemble algorithms for stream mining in multicore environments. With the aid of a formal framework, we demonstrate that mini-batching can significantly decrease the reuse distance (and the number of cache misses). Experiments on six different state-of-the-art ensemble algorithms applying four benchmark datasets with varied characteristics show speedups of up to 5X on 8-core processors. These benefits come at the expense of a small reduction in predictive performance.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Improving the performance of bagging ensembles for data streams through mini-batching

Cassales,

Gomes,

Bifet

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Qian et al . propose a novel multi‐thread grid‐based multi‐chunk multi‐level ensemble (GMCE) for data stream classification. In order to improve the learning efficiency, GMCE maps different raw data to multiple grids, represents the feature of the grid by the grid first‐order geometric center, and then classifies data based on the grid.…”

Section: Themes Of This Special Issuementioning

confidence: 99%

Embedded multicore computing and applications

Magoulès¹,

Hung

Jiang

et al. 2016

Concurrency and Computation

View full text Add to dashboard Cite

INTRODUCTIONIn the past few years, multicore and many core systems have been utilized to accelerate the research domains, including image processing, molecular dynamics, astrophysics, financial simulation, computer tomography, quantum chemistry, bioinformatics, and many others. Similarly, many programming models, such as OpenMP, OpenCL, and CUDA, have been released to aid programmers to develop a parallel computing application by launching the cores or threads in parallel. To efficiently launch cores or threads, software framework plays an important role to make cores or threads collaborate with each other. Also, an efficient parallel algorithm can help to improve the performance of applications, especially real-time applications. Many successful applications with multicore and many core systems through these parallel algorithms have unveiled many scientific results, which encourage application developers to adopt novel multicore and many core system computing technologies. This journal contains 10 papers, each paper covering one particular application and methodology. All of these papers not only provide novel ideas and state of-the-art technologies in the field but also stimulate future research for multicore computing and their applications. THEMES OF THIS SPECIAL ISSUEThis special issue contains research papers addressing the state-of-the-art in multicore systems and applications. A set of carefully selected works was invited based on the original presentations at the 4th International Workshop on Embedded Multicore Computing and Applications in conjunction with the 16th IEEE International Conference on High Performance and Communications, which was held in Paris, France, 20-22 August 2014 [1]. The manuscripts tackle research on different topics, including multicore systems, parallel input/output (IO), and graphics processing unit (GPU) applications. The set of accepted papers is briefly described in the remaining parts of this section. Parallel IO and parallel programming modelsMore than 50% of file operations in the storage systems are metadata operations. In the scenarios of read-more and write-less of massive small files, present distributed file systems suffer from the unsatisfying performance and scalability of metadata service because of random disk I/O during metadata operations. Duan et al.[2] propose a highly efficient metadata management for cloud storage systems. They introduce a concept of channel, which is an independent data storage pipe by binding an independent CPU core to each physical disk. In addition, a multi-channel fast key-value storage engine is proposed to provide the extremely efficient performance for the underlying storage service, which takes full advantages of multicore processors and parallel disk I/O.Nowadays, technology of multicore processor has not only taken more and more markets but also brought revolution and unavoidable collision to the programming personnel. In the past, the tenacious semiconductor problems of operating temperature and power consumption limited the pe...

show abstract

SubtStream: Online subtractive stream clustering algorithm

MİLLİ

Bulut

2022

Concurrency and Computation

View full text Add to dashboard Cite

Real-time stream data processing has gained high importance with the rapid rise of big data trends in different areas such as social media, finance, business, science, and bioinformatics. Stream data can be characterized as fast, unstable, and big data sets.Due to these properties of stream data, it cannot be processed effectively with traditional algorithms. Just like stream data processing, clustering is also a difficult task. However, researchers have attempted to classify stream data by modifying traditional algorithms or designing new ones. So, in previous studies, incremental methods were used for clustering the stream data. This paper highlights the need to develop an efficient real-time clustering algorithm for data streams in the presence of concept high drift and an adaptive algorithm for different dimensions. The proposed clustering algorithm, SubtStream, combines decremental (subtractive property) and incremental (additivity property) strategies to overcome the high drift. It also introduces a new dimension-based approach to adopt the dimension change. We use three radius parameters, Predefined User Parameter, Proactive Adaptive Parameter, and Reactive Adaptive Parameter, to achieve adaptability. The proposed method, SubtStream, showed better performance on synthetic and real data sets.

show abstract

Grid‐based high performance ensemble classification for evolving data stream

Cited by 5 publications

References 15 publications

Improving the performance of bagging ensembles for data streams through mini-batching

Improving the performance of bagging ensembles for data streams through mini-batching

Embedded multicore computing and applications

SubtStream: Online subtractive stream clustering algorithm

Contact Info

Product

Resources

About