Error-Correcting Output Codes - Coursework Example

COMPUTER SCIENCE AND INFORMATION TECHNOLOGY Introduction Error correcting output s (ECOC) are used in addressing varied problems in pattern recognition. For instance, designing of combined classifiers, ECOC provides diversified classifiers by means of dichotomies that result in accurate classification. The main use o f ECOC is to provide accurate classification of data in patterns, using dichotomies in computers. It has been used to organize classifiers in accurate data allocation and formation of patterns. The use of ECOC assists program developers to form accurate patterns of data, which recognize classifiers in data recognition. Moreover, its ability to reinforce classification accurately utilizes complex problems based on outputs. Therefore, complex problems can be fixed using ECOC by placing classification accurately and producing positive outputs of data. The classification of data requires proper use of ECOC, so that complex problems can be fixed effectively especially in structures- structured output prediction (Ngoc, Chong& Adam, 2011). Besides, the main motivation of using ECOC in computer science is to develop new structured methods for structured output prediction and concept of multiple classifier systems (MCS). The development of new structured methods and structured output prediction assists in giving output data that is accurate and reliable. Moreover, the prediction enables data users to have a prior knowledge of the figures they need to expect in the final output results. This enables for quality results of final data and data recognition that allows for an effective data classification of projects and general easy solving of complex problems. The concept of multiple classifiers systems is made easier using ECOC (Bhardwaj & Jain, 2010). The systems are easy to use in the computers through the implementation of ECOC in the development of multiple classifier systems. Therefore, Error Correcting Output Codes has been an essential computer technology and information development, since it is used in many applications for multiple data classification and designing of combined classifiers for effective data recognition. Consequently, varied problems in pattern recognition can be solved through the use ECOC, which give quality results of error correction by the use of the multiple classifiers systems (Assuncao & Costanazo, 2009). Error Correcting Output Codes Error-Correcting Output Codes were initiated in the field of pattern recognition for solving tribulations of multiple classes. The main idea for this development was to ignore solving multi-class problems openly but decompose into dichotomies. Multi-class patterns recognize problems that can be decomposed in limited quantity of two-class classification problems. Therefore, aggregated binary classifiers need to be able to recognize native set of predefined classes by isolating recognition tribulations into dichotomies. Moreover, the combination of binary classifiers makes it simple for nearest- neighbor rule to find the closest class into outputs of binary classifier combinations. Binary classifier combinations include one against one and one against all. One against one produces intuitive multi-class classifier where one binary classifier corresponds to each class. Besides, hypothesis that given objects belongs to selected class varies against membership of one of the others. Such approach flows in case of conflicting answers from classifiers that do not quite straightforward. On the other hand, the second approach one-against all strategy, uses Winner Takes All (WTA) rule. Every classifier trains on instances of separate class that belongs to the first class and all the other classes correspond to the second one. The final classification makes the basis of support functions through the maximization of the rule. Additionally, ECOC mentions proposed combination models of classifiers in different forms. The second approach, each sequence bits produced set binary classifiers in associated code words during learning. ECOC selects class with smallest hamming distance to its codeword. Besides, the method for structured prediction using ensemble classifiers composes using Error Correcting Output Codes. The presentation of multiple classifier methods for structured output prediction using Error Correcting Output Codes entails comparable computation time of other accurate algorithm. The classification of more complex structures provides the same time better results using Error Correcting Output Codes. Further experiments concern affects of structured patterns varieties on the calculation time. Dependency between coding methods, accuracy of classification and adequacy conventions method, differentiate major types of structures. Multi-class tribulations can be solved by joining numerous binary classifiers using error correcting output codes. According to Kukar (2007), Classes are programmed with fixed-length bit strings (codes) which divide problem into numerous sub problems through increasing distances between class codes. This assists in developing new solutions to numerous multi-class problems that may have been experienced in the learning process of binary classifiers. Therefore, the use of ECOC is useful in solving complex problems, pattern recognition of binary classifiers, provides diversified classifiers and general classification of systems (Kononenko& Kukar, 2007). It is fundamental for multi-classes to use ECOC in solving some of their tribulations based on binary classifiers since it gives them quality results that are accurate and relevant according to their data input. For instance, code length L, is transformed into L binary (two- class) sub problems. Each of binary sub problems merges original classes encoded with four in the first class and L in the second class up to the fifth class as illustrated below. The table is a sample for error correcting output code for five classes. Code is represented in matrix W i j. row I corresponds to one class and column j to one binary classifier. CLASSIFIERS CLASS 1 2 3 4 5 6 7 8 9 C1 1 1 1 4 1 4 1 1 1 C2 4 4 1 4 4 4 1 1 1 C3 4 4 1 1 1 1 4 4 4 C4 1 4 1 1 1 1 4 4 4 C5 1 1 4 1 1 4 1 4 1 From the table above, the code splits original problem into 9 two-class problems and each problem is solved by using its own classifier. The answers of all the 9 classifiers are merged into a bit string. The final classification is produced by finding row with the most similar obtained bit string in terms of differences. For instance, a bit string S= 14441141, differs with C1 in 8 bits,C2 and C3 in 3bits, C4 in 2 bits and finally C5 in 8 bits, therefore the correct answer becomes C4, since it differs with the original bit string with the minimal number. Therefore, the use of ECOC can be used to form patterns and structured output predictions in multiple classified systems. It recognizes and solves problems correctly. The use of ECOC is therefore important in all tribulations of data recognition and solving of problems. It splits major bit strings into sub-problems and analyses data to give the best possible answer to a problem. Moreover, in the presentation of ECOC, each row must be different to others rows as much as possible in order to increase number of bits that ECOC is able to correct. On the other hand, column resolution requires each bit code to stand for one two-class classification problem. The two-class problems must be uncorrelated. Therefore, all columns in the ECOC must be different for the realization of theses effect. Besides, the complementation of a column keeps two-class problem intact and therefore each column must be different from the other column that complements it. The level of correction is decrease by increasing difference between columns. Columns that consist of all ones and zeros are not useful since they represent trivial classification tribulations (Kononenko & Kukar, 2007). Multiple Classifier Systems The problem of designing combined classifiers is known as multiple classifier systems (MCSs) of classifier ensembles and it is grouped into three main issues:(1) on how to select topology of recognition system- the choosing of topology is usually chosen since it is intuitive and properly researched (2) on how to select classifier to the ensemble to make certain high diversity within group of chosen individual classifiers and finally, (3) the method of fusion need proposal on how to exploit strengths of chosen single classifiers (Ngoc, Chong& Adam, 2011). Focusing on the second method, the strategy for generating ensemble members improves diversity. Moreover, various components of MCS enforce classifier diversity using varied input data, such as various partitions of dataset and generating varied datasets by data separation. Additionally, bagging, cross-validated committee and boosting, all depend on classifiers that are trained on different inputs, which appear complementary (Buyya, 2010). Consequently, the use of classifiers with same input and output, trained based on varied models or versions, promote development of multiple classifiers systems. Moreover, using classifiers with different outputs such as individual classifiers, need to be trained in solving subset of multi-class problem like binary classifier, that is one class against the rest of the strategies and fusion methods need to recover the whole set of original classes. Zhou (2012) suggests that, all these are well known techniques’ of Error Correcting Output Codes (ECOC). The techniques’ are use to make date input and output manageable with the production of quality results that are relevant and comprehensive. The use of classifiers enables the possibility of solving different data inputs and outputs problems effectively using multiple classifier systems and Error Correcting Output Codes. Moreover, classifiers can be used with same data input and output, based on different models using multiple classifier systems and Error Correcting Output Codes (Santos, Gummadi & Rodrigues, 2009). Machine Learning Basics The measuring of machine learning basics with Error Correcting Output Code develops evaluations that have been invented in specific areas historically and dealt with through collection of data in sociology, medicine, information retrieval, biology and psychology (Beres, 2004). Many approaches have been adopted in the use of machine learning and often consider roughly equivalent methods to each other. Today, it has been shown alternative and often better measures invented like brier score and information score. In terms of regression problems there is fundamentally less confusion performance measures. The variations originate mostly from error means such as mean squared error, relative mean squared error and absolute error. All these are errors in regression problems that can be effectively corrected using Error Correcting Output Code (Ngoc, Chong& Adam, 2011). The correction of these errors contributes to vital changes in the performance measures of data input and output. The use of machine learning basics and Error Correcting Output Code yields high performance of data entry in measuring certain degree of data usage in computer systems. Moreover, the performance evaluation on small datasets, stratified cross validation and cross validation is used in Error Correcting Output Codes to give substantive results on the performance measurements. Besides, small datasets bootstrapping works gives better results than any other form of cross-validation that entails leave one out testing (Witten, & Frank, 2009). Error Correcting Output Codes and Machine Learning Basics The testing of fundamental performance variations, between machine learning algorithms develops more shaky grounds. Additionally, well known statistical tests have been used to enable major corrections on this purpose. Through research, it has been shown that certain properties of experimental results acquired from machine learning algorithms, the results are treated independently. It is incorrect to use basic tests to solve problems of this purpose although several possible solutions have been projected. The use of Bonferroni correction as Error Correcting Output Code, in machine learning basics seriously reduces power of underlying tests (Ngoc, Chong& Adam, 2011). However, many authors have proposed different correcting tests such as t-test. The advocating of the use of non-parametric tests, especially improved Friedman test has been proposed by many programmers since it produces quality results that are useful in the development of machine language and comprehensions of data entry. Combining answers of several classifiers, gives the best-isolated answer or some kind of a more weighted vote. Consequently, advanced general methods combine classifiers that entail boosting, random forests and bagging (Murata, et al., 2010). Numerous binary classifiers have been combined through error correcting output codes in order to solve multi-class tribulations. Moreover, transductive approach suggests that in order to solve tribulations one needs to avoid solving problems that are more general as intermediate. Chong& Adam (2011) argue that this application has further been applied and refined by many programmers in solving problems using Error Correcting Output Code in machine learning basics. The use of different approaches in solving general tribulations of data entry makes it easier for programmers to input data and manipulate it to quality standards of acceptable results. More to the point, the use of different approaches facilitates pattern recognition, diversified classification and structured output predictions data entry in computer systems through Error Correcting Output Codes (Patel, Ranababu & Sheth, 2009). The learning of different machine basics accelerates developments in artificial intelligence of humans and other important aspects. The approaches increases artificial intelligence in different areas and fields such biology, psychology and medicine. Increased intelligence in these fields increases the ability of research in the same field and hence develops quality results in the performance of duties in the respective fields. The use of Error Correcting Output Code, allows for data entry that gives best results through proper formation of patterns within the system (Schneidewind, United States, & Naval Postgraduate School (U.S.) 2004). Conclusion The use of Error Correcting Output Codes and machine learning basics in data input and output produces varied results. The results need proper manipulations using ECOC to give accurate and reliable quality results, which has fundamental objectives in different fields of science and psychology. ECOC, works through addressing pattern recognition problems in data input, produces diversified classifiers, uses structured output predictions and multiple classified systems in data manipulations to produce quality and accurate results that are fundamental in varied fields. Moreover, the use of ECOC enables increases artificial intelligence since data input and output produced are accurate and reliable in varied forms. The use of ECOC has enabled major different fields to prosper in terms in research and intelligence in the fields. It is important to use ECOC since it generates multiple classifiers systems that yield accurate results. Reference Assuncao,M.D.,Costanazo,A(2009): Evaluating The Cost –Benefit Of Using Cloud Computing To Extend The Capacity Of Clusters, In: 18th ACM International Symposium On High Performance Distributed Computing,P.P 141-150,ACM Press, New York Bhardwaj,S.,Jain,L.,Jain,S., (2010):Cloud Computing:A Study Of Infrastructure As A Service (IAAS). Journal Of Engineering And Information Technology 2(1), 60-63 Beres, E. (2004). Blind Channel Estimation For Orthogonal Space-Time Block Codes In MISO System Buyya,R(2010): Service Level Agreement (SLA) In Utility Computing Systems. Technical Report, CLOUDS-TR-2010-5, Cloud Computing And Distributed Systems Laboratory Santos,N.,Gummadi,K.P,.Rodrigues,R(2009):Towards Trusted Cloud Computing.In:Work-Shop On Hot Topics In Cloud Computing, USENIX Patel,P.,Ranababu,A.,Sheth,A(2009):Service Level Agreement In Cloud Computing.In: Cloud Workshops At OOPSLA Murata,Y.E.,Higashida,R.,Kobayashi,M.,Cybersci,H(2010): A History-Based Job Scheduling Mechanism For The Vector Computing Cloud.In: 10th Annual Symposium On Applications & The Internet,Pp.125-128.Ieeepress, New York Ngoc Thanh Nguyen, Chong-Gun Kim, Adam Janiak(2011): Intelligent Information And Database Systems: Third International Conference, ACIIDS 2011, Daegu, Korea, April 20-22, 2011, Proceedings. Springer,580 Pages Kononenko.I, M Kukar(2007): Machine Learning And Data Mining. Elsevier;480 Pages Witten, I. H., & Frank, E. (2009). Data Mining: Practical Machine Learning Tools And Techniques With Java Implementations. San Francisco, Calif: Morgan Kaufmann. Schneidewind, N. F., United States., & Naval Postgraduate School (U.S.). (2004). Analysis Of Error Processes In Computer Software. Monterey, California: Naval Postgraduate School Zhou, Z.-H. (2012). Ensemble Methods: Foundations And Algorithms. Boca Raton, FL: Taylor & Francis. Read More