A. Baraldi and E. Alpaydin. "Constructive feedforward art clustering networks -- Part II". IEEE Transactions on Neural Networks, Vol. 13, No. 3, May 2002.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....classification problem, where each new perception is classified as belonging to a known category after the algorithm has been trained on many examples of all the different classes. We used a novelty filter known as the Grow When Required (GWR) network [15] see also the similar FaSART algorithm [2, 1]) to learn a model of a training environment. The GWR network is a self organising topology preserving network that can dynamically create and destroy network nodes, allowing it to model dynamic datasets on line. When the robot explores a test environment, any perceptions that the novelty ....

A. Baraldi and E. Alpaydin. Constructive feedforward ART clustering networks - part II. IEEE Transactions on Neural Networks, 13(3):662-677, 2002.

.... Radial Basis Functions Networks (SL) 31] Growing Cell Structures (GCS, SL and UL) 32] Growing Neural Gas (GNG, SL and UL) 33, 34] that take Neural Gas [35] UL) as a starting point, Fuzzy ARTMAP (SL) 36] that derives from the Adaptive Resonance Theory (ART) of Grossberg [37] FOSART [29,38] and the Competitive Agglomeration Algorithms [2, 25] In this paper the Fully Automatic Clustering System (FACS) is presented. It is a VQ CA iterative algorithm whose aim is, given a data set and a target error e T , to nd a codebook that approximates the input data set with an error less than ....

....a drawback of FACS, but it derives from the nature of the algorithm itself. However, there are applications where the identi cation of the outliers is essential in order to avoid their in uence on the nal result. In such cases, FACS can still be used, provided a noise category removal mechanism [38] is added for treating the outliers. Even if it is outside the scope of this paper, we wish to cite some methods commonly adopted in literature that, opportunely readapted, could represent the desired modi cations. For example, a simple mechanism lets the algorithm run with the desired value of e ....

[Article contains additional citation context not shown here]

A. Baraldi and E. Alpaydin, \Constructive feed-forward ART clustering networks - Part I," IEEE Transactions on Neural Networks, in press.

....codewords) are highlighted in Figs 1(c) and 1(d) respectively. It is evident that, in the latter case, the degree of approximation is better that in the former. In spite of the di erences outlined here, it is possible to demonstrate that, in many cases, CA and VQ are, practically, equivalent [27 29]. Summarizing, we can say that, often, from an operative point of view, the two approaches roughly execute the same operations: grouping data into a certain number of groups so that a loss (or error) function is minimized. In literature, several techniques for VQ CA exist where, as the algorithm ....

.... Radial Basis Functions Networks (SL) 31] Growing Cell Structures (GCS, SL and UL) 32] Growing Neural Gas (GNG, SL and UL) 33, 34] that take Neural Gas [35] UL) as a starting point, Fuzzy ARTMAP (SL) 36] that derives from the Adaptive Resonance Theory (ART) of Grossberg [37] FOSART [29,38] and the Competitive Agglomeration Algorithms [2, 25] In this paper the Fully Automatic Clustering System (FACS) is presented. It is a VQ CA iterative algorithm whose aim is, given a data set and a target error e T , to nd a codebook that approximates the input data set with an error less than ....

[Article contains additional citation context not shown here]

A. Baraldi and E. Alpaydin, \Constructive feed-forward ART clustering networks - Part II," IEEE Transactions on Neural Networks, in press.

.... Radial Basis Functions Networks (SL) 62] Growing Cell Structures (GCS, SL and UL) 63] Growing Neural Gas (GNG, SL and UL) 64, 65] that take Neural Gas [66] UL) as a starting point, Fuzzy ARTMAP (SL) 67] that derives from the Adaptive Resonance Theory (ART) of Grossberg [68] FOSART [48, 69] and the Competitive Agglomeration Algorithms [14, 45] In chapter 6, the Fully Automatic Clustering System (FACS) is presented. It is a VQ CA iterative algorithm whose aim is, given a data set and a target error e T , to nd a codebook that approximates the input data set with an error less than ....

....a drawback of FACS, but it derives from the nature of the algorithm itself. However, there are applications where the identi cation of the outliers is essential in order to avoid their in uence on the nal result. In such cases, FACS can still be used, provided a noise category removal mechanism [69] is added for treating the outliers. Even if it is outside the scope of this work, we wish to cite some methods commonly adopted in literature that, opportunely readapted, could represent the desired modi cations. For example, a simple mechanism lets the algorithm run with the desired value of e ....

[Article contains additional citation context not shown here]

A. Baraldi and E. Alpaydin, \Constructive feed-forward ART clustering networks - Part I," IEEE Transaction on Neural Networks, in press.

....codewords) are highlighted in Figs 1.1(c) and 1.1(d) respectively. It is evident that, in the latter case, the degree of approximation is better than in the former. In spite of the di erences outlined here, it is possible to demonstrate that, in many cases, CA and VQ are, practically, equivalent [46 48]. Summarizing, we can say that, often, from an operative point of view, the two approaches roughly execute the same operations: grouping data into a certain number of groups so that a loss (or error) function is minimized. In the remainder of the thesis, we will use a symbology and terminology ....

.... Radial Basis Functions Networks (SL) 62] Growing Cell Structures (GCS, SL and UL) 63] Growing Neural Gas (GNG, SL and UL) 64, 65] that take Neural Gas [66] UL) as a starting point, Fuzzy ARTMAP (SL) 67] that derives from the Adaptive Resonance Theory (ART) of Grossberg [68] FOSART [48, 69] and the Competitive Agglomeration Algorithms [14, 45] In chapter 6, the Fully Automatic Clustering System (FACS) is presented. It is a VQ CA iterative algorithm whose aim is, given a data set and a target error e T , to nd a codebook that approximates the input data set with an error less than ....

[Article contains additional citation context not shown here]

A. Baraldi and E. Alpaydin, \Constructive feed-forward ART clustering networks - Part II," IEEE Transaction on Neural Networks, in press.

No context found.

A. Baraldi and E. Alpaydin. "Constructive feedforward art clustering networks -- Part II". IEEE Transactions on Neural Networks, Vol. 13, No. 3, May 2002.

No context found.

Baraldi, A. and Alpaydin, E. (2002). Constructive feedforward ART clustering networks---part II. IEEE Transactions on Neural Networks, 13(3):662--677.

No context found.

A. Baraldi and E. Alpaydm, "Constructive feedforward ART clustering networks---Part I," IEEE Trans. Neural Networks, vol. 13, pp. 645--661, 2002.

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