T. Marill and D. Green, "On the effectiveness of receptors in recognition systems," IEEE Transactions on Information Theory, vol. 9, pp. 11--17, 1963.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....densities, for its connection with information theory, is the Kullback Leibler distance, p(x[Ci) r x (6) KL(Ci, Cj) p(xlCi) log p(xiCD where 1 i, j K. The asymmetry of Kullback Leibler motivates the symmetric measure of divergence, since long ago used for feature selection [21], defined as Dij = D(Ci, Cj) KL(Ci, Cj) KL(Cj, Ci) 7) Besides being symmetric, divergence is zero between a distribution and itself, always positive, monotonic on the number of features and provides an upper bound for the classification error [22] The two main drawbacks of divergence are ....

Marill, T., Green, D.: On the effectiveness of receptors in recognition systems. IEEE Trans. on Information Theory O (1963) 1-17

....start from a small subset of features and repeatedly add additional features to it while backward features selection start from the set of all features and repeatedly remove features from it. A simple sequential method like Sequential Forward Selection (SFS) or Sequential Backward Selection (SBS) [10] adds (deletes in the case of SBS) one feature at a time. More sophisticated algorithms including the Sequential Floating Search methods SFFS and SFBS [12] use backtracking as long as they find improvement in the criterion function compared to the previous feature set of the same size. In most ....

....of the same size. In most cases SFFS and SFBS, perform better than the straight sequential searches. However in our experiments and for simplicity we have used the simple Sequential Backward Selection (SBS) 3. 1 Sequential Backward Selection (SBS) algorithm SBS, introduced by Marill and Green [10], works in top down fashion. SBS algorithm starts from the full set of features and sequentially deletes features one by one until no improvement is achieved in the selected evaluation function. Suppose k measurements have been removed from the set of measurements Y= y j j=1, D , forming a ....

T. Marill and D.M. Green, "On the effectiveness of receptors in recognition systems", IEEE Trans. Inform. Theory V. 9 , pp 11-17, 1963

....feasible. Feature selection is also related to four other areas of research: dimensionality reduction [24] space partitioning [17] feature extraction and decision trees [21] Many algorithms have been proposed for feature selection, from simple algorithms like Sequential Forward Selection (SFS) [18] to more complex algorithms such as neural net prunning [6] and genetic selection [22] Surveys of feature selection algorithms are given by Kittler [12] Siedlecki and Sklansky [22] and Bins [1] For this work, the most relevant algorithms are: Relief, proposed by Kira and Rendell [11] in 1992 ....

T. Marill and D.M. Green, "On the Effectiveness of Receptors in Recognition Systems", IEEE transactions on Information Theory, 9:11-17, 1963.

....feature subsets because it is the relative accuracy that matters most. 1. 5 RELATED WORK The pattern recognition and statistics literature offers many filter approaches for feature subset selection (Devijver and Kittler, 1982; Neter et al. 1990) Sequential backward elimination was introduced by Marill and Green (1963). Most machine learning induction algorithms do not obey the monotonic restrictions that underlie much of the early work in statistics and pattern recognition, and they are applied to databases with a large number of features, so they require special heuristic methods. Some more recent work in ....

Marill, T. and Green, D. M. (1963). On the effectiveness of receptors in recognition systems. IEEE Transactions on Information Theory, 9:11--17.

....can be used, the search is usually exponential, and when there are more than 30 features, suboptimal methods are used. Searching in the space of feature subsets has been studied for many years. Sequential backward elimination, sometimes called sequential backward selection, was introduced by Marill Green (1963). Kittler (1978) generalized the different variants including forward methods, stepwise methods, and plus take away r. Cover Campenhout (1977) showed that even for multivariate normally distributed features, no hill climbing procedure that uses a monotonic measure and that selects one ....

Marill, T. & Green, D. M. (1963), "On the effectiveness of receptors in recognition systems", IEEE Transactions on Information Theory 9pp. 11--17.

....to improve the comprehensibility of extracted knowledge [KJ96] Machine learning has taken inspiration and borrowed from both pattern recognition and statistics. For example, the heuristic search technique sequential backward elimination (section 3. 3) was first introduced by Marill and Green [MG63] Kittler [Kit78] introduced different variants, including a forward method and a stepwise method. The use of 26 cross validation for estimating the accuracy of a feature subset which has become the backbone of the wrapper method in machine learning was suggested by Allen [All74] and applied ....

T. Marill and D. M. Green. On the effectiveness of receptors in recognition systems. IEEE Transactions on Information Theory, 9:11--17, 1963. 175

....and its use within the computational learning community has become quite widespread within the last few years. In statistics, research on feature selection has focused primarily on selecting a subset of features within linear regression. Techniques developed include sequential backward selection [20], branch bound [22] and search algorithms [23, 25] A 1993 meeting of the Society of AI and Statistics was dedicated to papers on Selecting Models from Data [7] and contains a large number of papers on feature selection. This statistical approach to subset selection shares many principles with ....

T. Marill and D. Green. On the effectiveness of receptors in recognition systems. IEEE Trans. on Information Theory, 9:11--17, 1963.

....bound can be used, the search is usually exponential, and when there are more than 30 features, suboptimal methods are used. Searching in the space of feature subsets has been studied for many years. Sequential backward elimination, sometimes called sequential backward selection, was introduced by Marill Green (1963). Kittler (1978) generalized the different variants including forward methods, stepwise methods, and plus take away r. Cover Campenhout (1977) showed that even for multivariate normally distributed features, no hill climbing procedure that uses a monotonic measure and that selects one ....

Marill, T. & Green, D. M. (1963), "On the effectiveness of receptors in recognition systems", IEEE Transactions on Information Theory 9, 11--17.

....30,000,000 combinations if exhaustive search is used Clearly, exhaustive search is impractical even for problems of moderate size. Rather than using exhaustive search, we consider two methods used in pattern recognition for large scale feature selection sequential backward selection (SBS) [9] and branch and bound (B B) 10] The B B algorithm is optimal and more efficient than exhaustive search, but it is significantly more complex than the SBS algorithm. Even though the SBS algorithm is suboptimal, in general it still performs quite well, as we will demonstrate in Section IV, and may ....

T. Marill and D. M. Green, "On the effectiveness of receptors in recognition systems," IEEE Transactions on Information Theory, vol. 9, pp. 11--17, 1963.

....and its use within the computational learning community has become quite widespread within the last few years. In statistics, research on feature selection has focused primarily on selecting a subset of features within linear regression. Techniques developed include sequential backward selection [Marill63], branch bound [Narendra77, Xu89] and search algorithms [Siedlecki88] A 1993 meeting of the Society of AI and Statistics was dedicated to papers on Selecting Models from Data [Cheeseman94] and contains a large number of papers on feature selection. This statistical approach to subset selection ....

Marill, T. and D. Green (1963). On the effectiveness of receptors in recognition systems. IEEE Trans. on Information Theory, 9:11--17.

....feature subsets because it is the relative accuracy that matters most. 1. 5 RELATED WORK The pattern recognition and statistics literature offers many filter approaches for feature subset selection (Devijver and Kittler, 1982; Neter et al. 1990) Sequential backward elimination was introduced by Marill and Green (1963). Most machine learning induction algorithms do not obey the monotonic restrictions that underlie much of the early work in statistics and pattern recognition, and they are applied to databases with a large number of features, so they require special heuristic methods. More recent work in feature ....

Marill, T. and Green, D. M. (1963). On the effectiveness of receptors in recognition systems. IEEE Transactions on Information Theory, 9:11--17.

....and its use within the computational learning community has become quite widespread within the last few years. In statistics, research on feature selection has focused primarily on selecting a subset of features within linear regression. Techniques developed include sequential backward selection (Marill and Green, 1963), branch bound (Narendra and Fukunaga, 1977) and search algorithms (Siedlecki and Sklansky, 1988) Feature selection has received considerable attention in the last few years within the computational learning community, using both filter based and wrapper based approaches (John et al. 1994) A ....

Marill, T. and Green, D. (1963). On the effectiveness of receptors in recognition systems. IEEE Trans. on Information Theory, 9:11--17.

.... A number of class separability measures have been introduced in the pattern recognition literature[9, 17, 46, 74, 47, 31] The criteria used by all of these methods, however, can be categorised into five different classes: probability of error[130] interclass distance[125] probability distance[103, 9, 17, 105, 104], probability dependence[129] and entropy measures[91, 2] The probability of error as a separability measure is useful where the performance of a system can be evaluated by it. This is, however, not practical since the computation of the probability of error is not feasible. The interclass ....

T. Marill and D.M. Green. On the effectiveness of receptors in recognition systems. IEEE Trans. Inform. Theory, 9:11--17, 1963.

....a very high computational cost. 6 Related Work Feature selection has been widely used in statistics and pattern recognition, with research in this area being focused primarily on selecting a subset of features within linear regression. Techniques developed include sequential backward selection (Marill Green, 1963), branch and bound (Narendra Fukunaga, 1977) best first (Xu et al. 1989) and beam search as well as bidirectional search (Siedlecki Sklansky, 1988) A recent meeting of the Society of AI and Statistics was dedicated to papers on Selecting Models from Data (Cheeseman Oldford, 1994) and ....

Marill, T. & Green, D. (1963). On the effectiveness of receptors in recognition systems. IEEE Trans. on Information Theory, 9:11--17.

....used, the search is usually exponential, and when there are more than 30 or 40 features, heuristic methods need to be used. Searching in the space of feature subsets has been studied for many years. Sequential backward elimination, sometimes called sequential backward selection, was introduced by Marill Green (1963). Kittler (1978) generalized the different variants including forward methods, stepwise methods, and plus take away r. Cover Campenhout (1977) showed that even for multivariate normally distributed features, no hill climbing procedure that uses a monotonic measure and that selects one ....

Marill, T. & Green, D. M. (1963), "On the effectiveness of receptors in recognition systems", IEEE Transactions on Information Theory 9 pp. 11--17.

....recognition (Devijver Kittler 1982; Ben Bassat 1982) have investigated the feature subset selection problem for decades, but most work has concentrated on subset selection using linear regression. Sequential backward elimination, sometimes called sequential backward selection, was introduced in Marill Green (1963). Kittler generalized the different variants including forward methods, stepwise methods, and plus take away r. Branch and bound algorithms were introduced by Narendra Fukunaga (1977) Finally, more recent papers attempt to use AI techniques, such as beam search and bidirectional search ....

Marill, T., and Green, D. M. 1963. On the effectiveness of receptors in recognition systems. IEEE Transactions on Information Theory 9:11--17.

No context found.

T. Marill and D. Green, "On the effectiveness of receptors in recognition systems," IEEE Transactions on Information Theory, vol. 9, pp. 11--17, 1963.

No context found.

T.Marill and D.M.Green, "On the Effectiveness of Receptors in Recognition Systems", IEEE Trans., IT-9,1963, pp.11-27

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC