Generally, big data computing deals with massive and high dimensional data such as DNA microrray data, financial data, medical imagery, satellite imagery and hyperspectral imagery. Therefore, big data computing needs advanced technologies or methods to solve the issues of computational time to extract valuable information without information loss. In this context, generally, Machine Learning (ML) algorithms have been considered to learn and find useful and valuable information from large value of data. However, ML algorithms such as Neural Networks are computationally expensive, and typically the central processing unit (CPU) is unable to cope with these requirements. Thus, we need high performance computer to execute faster solutions such Graphical Processing Unit (GPU). GPUs provide remarkable performance gains compared to CPUs. The GPU is relatively inexpensive with affordable price, availability and scalability. Since 2006, NVIDIA provides simplification of the GPU programming model with the Compute Unified Device Architecture (CUDA), which supports for accessible programming interfaces and industry-standard languages, such as C and C++. Since then, General Purpose Graphical Processing Unit (GPGPU) using ML algorithms are applied on various applications; including signal and image pattern classification in biomedical area. The importance of fast analysis of detecting cancer or non-cancer becomes the motivation of this study. Accordingly , we proposed machine learning framework and analytics of Self Organizing Map (SOM) and Multiple Back Propagation (MBP) for big biomedical data classification problems. Big data such as gene expression datasets are executed on high performance computer and Fermi architecture graphical hardware. Based on the experiment, MBP and SOM with GPU - Tesla generates faster computing times than high performance computer with feasible results in terms of speed performance.

http://home.ijasca.com/data/documents/IJASCA_060216_Final_Shasha.pdf

Machine Learning Big Data Framework and Analytics for Big Data Problems

This paper describes the paraphrase generation based on n-gram approach. N-grams are relevant words of text document that can be applied for a range of Natural Language Processing (NLP) applications. The candidate paraphrases are generated based on trigrams approach. The reference paraphrases (keyphrases) are the set of relevant paraphrases, which acts like training data set for generating candidate paraphrases. The task of paraphrase generation is similar to machine translation; hence we used machine translation evaluation metrics. R-precision evaluation metric is used to find the number of common words between candidate and reference paraphrases.

N-gram based paraphrase generator from large text document

Indexing n-gram phrases from text has many practic al applications. Plagiarism detection, comparison of DNA of sequence or spam detection. In this paper we describe several data structures l ike hash table or B+ tree that could store n-grams for searching. We perform tests that shows their advantages and disadvantages. One of neglected data structure for this purpose, ternary search tree, is deeply described and two performanc e improvements are proposed.

/pdf/efficient-in-memory-data-structures-for-n-grams-indexing-2r0549x0fr.pdf

Efficient in-memory data structures for n-grams indexing

Generally, big data computing deals with massive and high-dimensional data such as DNA microarray data, financial data, medical imagery, satellite imagery, and hyperspectral imagery. Therefore, big data computing needs advanced technologies or methods to solve the issues of computational time to extract valuable information without information loss. In this context, generally, machine learning (ML) algorithms have been considered to learn and find useful and valuable information from large value of data. However, ML algorithms such as neural networks are computationally expensive, and typically, the central processing unit (CPU) is unable to cope with these requirements. Thus, we need a high-performance computer to execute faster solutions such graphics processing unit (GPU). GPUs provide remarkable performance gains compared to CPUs. The GPU is relatively inexpensive with affordable price, availability, and scalability. Since 2006, NVIDIA provides simplification of the GPU programming model with the Compute Unified Device Architecture (CUDA), which supports for accessible programming interfaces and industry-standard languages, such as C and C++. Since then, general-purpose graphics processing unit (GPGPU) using ML algorithms are applied on various applications, including signal and image pattern classification in biomedical area. The importance of fast analysis of detecting cancer or non-cancer becomes the motivation of this study. Accordingly, we proposed soft computing methods, self-organizing map (SOM) and multiple back-propagation (MBP) for big data, particularly on biomedical classification problems. Big data such as gene expression datasets are executed on high-performance computer and Fermi architecture graphics hardware. Based on the experiment, MBP and SOM with GPU-Tesla generate faster computing times than high-performance computer with feasible results in terms of speed and classification performance.

Soft Computing Methods for Big Data Problems

Every application nowadays produces huge amounts of log data which can give critical insights if properly monitored. The biggest problem in monitoring such large amounts of log data by using the existing techniques is that they are compute intensive, and require a lot of effort. And also usually log data is unformatted and may contain some redundant data as well that needs to be removed. This critical information is more valuable, a potential predictive analytics improves the proactive control of the application. Data analytics paradigm shows new ways of handing this critical information by providing various technologies to handle this in an effective way. Visualization is used to convey the findings from data analytics in an easily understandable way .Due to the huge size of data that has to be analyzed, a highly scalable approach is required which provides additional optimization. In this paper an approach for data analytics with the combination of cloud computing is proposed and demonstrated.

Log analytics on cloud using pattern recognition a practical perspective to cloud based approach

N-grams are applied in some applications searching in text documents, especially in cases when one must work with phrases, e.g. in plagiarism detection. N-gram is a sequence of n terms (or generally tokens) from a document. We get a set of n-grams by moving a floating window from the begin to the end of the document. During the extraction we must remove duplicate n-grams and we must store additional values to each n-gram type, e.g. n-gram type frequency for each document and so on, it depends on a query model used. Previous works utilize a sorting algorithm to compute the n-gram frequency. These approaches must handle a high number of the same n-grams resulting in high time and space overhead. Moreover, these techniques are often main-memory only, it means they must be executed for small or middle size collections. In this paper, we show an index-based method to the n-gram extraction for large collections. This method utilizes common data structures like B+-tree and Hash table. We show the scalability of our method by presenting experiments with the gigabytes collection.

Index-based n-gram extraction from large document collections

This paper describes a utilization of the Self Organizing Map (SOM) method for the analysis of power outage data. SOM, to be already used in many fields, is based on the Kohonen self-organizing neural network and it is known to capture underlying concepts. We apply this method for a unified database of power outages to be collected for several years in the Czech Republic. The most significant attributes are selected from the database and records are used for the training of the SOM. We utilize our previously introduced application EDAS (Electrical Data Analysis using SOM) for the visualization, understanding, and analysis of the trained SOM. Because of performance issues in the previous introduced approaches, we implement our SOM on GPU environment and compare this method with previous solutions in this article.

Efficient Computation of SOM for Outage Database

Cache of a persistent data structure represents an important part, which significantly influence its efficiency. Cache consists from an array of main memory blocks (caled cache nodes) with a constant size. Cache nodes buffer the data structure nodes hence they can be accessed quickly. However, data structure nodes do not usually fully utilize the whole main memory block. Therefore, the constant cache node size the waste of the main memory. In this article, we propose the solution where the cache consists from the cache nodes with a different size. Cache is split into several cache rows. The data structure nodes with the similar size are stored in one cache row. Such a solution has two problems: (1) that the some cache row has to replace the nodes too often during the querying (cache row contention), and (2) that the data structure nodes has to be moved between the cache rows during the insert and the delete operation. In our experimental section we show that the effect of the cache row contention is minimized if we set the cache row size according to the data structure node distribution. We also discuss a solution of the node cache moves problem.

/pdf/flexible-cache-for-database-management-systems-4y5ac3zvcx.pdf

David Bednar

Papers

Index-based n-gram extraction from large document collections

Efficient Computation of SOM for Outage Database

Flexible Cache for Database Management Systems.