Utgoff, P. E. (1989). Perceptron trees: A case study in hybrid concept representations. Connection Science, 1, 377-391.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....(2 D) top view representation. Fig.1(a) represents an example of a pipeline image, while Fig. 1(b) and (c) shows the same image after luminosity compensation and 2 D top view transformation, respectively. The input image is divided into regions (called macro pixels) and a neural tree (NT) [16] [19] is used to classify each region into different object classes (e.g. pipeline, sea bottom and anodes) Each macro pixel is then analyzed according to geometric constraints: macro pixels with uncertain classification are divided into four parts and re classified. The process is iterated ....

P. E. Utgoff, "Perceptron tree: A case study in hybrid concept representation, " in Proc. 7th Nat. Conf. Artificial Intelligence, 1988, pp. 601--605.

.... methods is given in [23] Behnke and Karayiannis use competitive learning to form a competitive decision tree architecture named CNET [2] A hybrid form which contains neural networks at the leaves of the tree and univariate nodes in the nonleaf nodes of the tree was proposed by Utgoff [22]. IV. PRUNING A greedy algorithm is a local search method where at each step, one tries to make the best decision and proceeds to the next decision, never backtracking and reevaluating a decision after it has been made. Similarly in decision tree induction, once a decision node is fixed, it ....

P. E. Utgoff, "Perceptron trees: A case study in hybrid concept representations, " Connection Sci., vol. 1, pp. 377--391, 1989.

....Such a criterion makes it possible to vary the model complexity across the signal space. One of the more flexible approaches to model representation at various levels is a tree structure. This approach has been used for similar purposes under names such as regression and neural trees, see e.g. (Utgoff, 1989). Gained knowledge, i.e. experienced stimulus response pairs associated with a high reward, is represented by a behavior probability density function. This distribution is estimated by partitioning the set of experienced decisions, which are seen as samples from the behavior distribution, and ....

Utgoff, P. E. (1989). Perceptron trees: A case study in hybrid concept representations. Connection Science, 1:377--391.

....decomposed into a sequence of simpler local classification rules. Decision trees can be straightforwardly combined with neural networks since neural network modules can be used to make the decisions at the branch nodes in the tree. For this reason a number of neural tree models have been proposed [121, 132, 19, 128, 36, 119, 127, 87]. For neural computing the neural tree architecture has the advantage that it can be readily implemented in hardware because of its recursive nature. At a branch tree node in a classification tree the outcome of the decision causes the loading of one of two sets of weights (Figure 20) This ....

P.E. Utgoff. Perceptron trees: a case study in hybrid concept representations. In Proceedings of the Seventh AAAI National Conference on Artificial Intelligence, pages 601--606. Morgan Kaufmann: San Mateo, CA, 1988.

....such that it moves beyond the training set, to not only correctly classify objects from the training set but other unseen objects as well. There has been a plethora of research in the area of inducing a classifier for the above induction problem using the Decision Tree formalism (Quinlan, 1983) (Utgoff, 1988), Utgoff, 1989) However, there has been very little work in the area of directly inducing DNF representations for concepts. In fact, most current work in this area (Pagallo, 1989) has tended to use the decision tree formalism as the basis. 3.2 Computational Learning There has been considerable ....

Utgoff, P. E. (1988). Perceptron trees: A case study in hybrid concept representations. Proceedings of the Seventh National Conference on Artificial Intelligence (pp. 601-606). Saint Paul, MN: Morgan Kaufmann.

....example, it has been shown that the orthogonality of class boundaries is often a too strong constraint that might deeply affect the accuracy and the simplicity of the decision surfaces drawn in the description space. Techniques like oblique decision trees [Murthy and al. 1994] or perceptron trees [Utgoff, 1988] remove that bias. IBL [Aha and al. 1991] and Prototype based techniques [Kohonen, 1990; Decaestecker, 1993] implicitly induce Piecewise linear boundaries. Other techniques such as Typical Instance Based Learning [Zhang, 1992] or Numerical Flexible Decision Trees [Van de Merckt, 1992] produce ....

Utgoff P.E. (1988) Perceptron Trees: A case Study in Hybrid Concepts Representations, Proceedings of AAAI-88.

....PERFORMANCE OF MULTI RULE SET APPROACH FOR DIFFERENT NUMBER OF RULE SETS. Application 5 Rule sets 10 Rule sets 20 Rule sets GL 0.6215 0.6206 0.6216 SBD 0.5311 0.5490 0.5665 VII. COMPARISON WITH NEURAL NETWORKS We also compared our learning environment with a neural tree learning environment [19], which is a multilayered, tree structured decision making scheme in which each node consists of a neural network. Neural networks associated with intermediate nodes decide which path in the tree to follow, whereas neural networks associated with leafs make final decisions. If a single neural ....

P. Utgoff, "Perceptron trees: a case study in hybrid concept representation," in Proc. 7th National Conf. Artificial Intelligence, Seattle, 1988, pp. 601-605.

....algorithm (section 3.1) has been implemented in a program called LCSP. It consists of a problem solver, an interface to CINDI, and an interface for learning algorithms. Several learning algorithms are available, including LTU s [Nilsson, 1965] ID5R [Utgoff, 1989] the Perceptron Tree algorithm [Utgoff, 1988], the DNC algorithm [Ash, 1989] and the Taylor algorithm for learning polynomial evaluation functions 10 ; each of these can be used with LCSP and CINDI. This suite of programs enables rapid testing of any change to the theory of knowledge based feature generation. 4.2 Expected Results The ....

Utgoff, P. E. (1988). Perceptron trees: A case study in hybrid concept representations. Proceedings of the Seventh National Conference on Artificial Intelligence (pp. 601-606). Saint Paul, MN: Morgan Kaufmann.

....by backpropagation or a similar algorithm. The main advantage of this method is the time and effort saved in avoiding the exhaustive testing of different neural network topologies to obtain an acceptable solution. It is also possible to embed neural elements within a decision tree architecture [94]. These Perceptron trees integrate linear threshold units (LTU) within each leaf node of the decision tree. A similar technique exists that uses sigmoid functions in place of the LTUs and is called an entropy net [77] The tree based neural net (TBNN) system of Ivanova and Kubat [46] generalizes ....

P. E. Utgoff. Perceptron trees: a case study in hybrid concept representations, Connection Science, 1(4):377-391, 1989.

....Note that if some performance criteria is available it is possible to stop the model complexity from growing larger than necessary. This makes it possible to vary the model complexity across the signal space. These estimation procedures resembles to the use of regression and neural trees [6, 8]. A major advantage with tree strucures is that it makes it possible to benefit from the signal models being local by using branch and bound techniques. No effort has to be spent on models that are invalid in the current context. Instead the search proceeds down the most promising branch. A well ....

Paul E. Utgoff. Perceptron trees: A case study in hybrid concept representations. Connection Science, 1:377--391, 1989.

....structured and which, considering the network topology of nodes and propagating links, are indeed trees. The construction of these neural trees however is not done in a hierarchical way by a neural training 3 HIERARCHICAL MODELS mechanism. Tree structured MLP networks like the Perceptron Trees [Utg88] are quite comparable to decision trees (see below) as they use an attribute test at each decision node. But unlike in decision trees, now the terminal nodes are not class assignment nodes, but Linear Treshold Units (LTU) which divide the current subspace by a hyperplane. Thus, the terminal nodes ....

P. Utgoff. Perceptron trees: A case study in hybrid concept representations. In Proc. of the Nat. Conf. on AI, pages 601--606, St. Paul, MN, 1988.

....error accumulation from the smaller parent SOMs to the lowest child SOM. However, the effect of the Probing Algorithm is limited by the (in general) non optimal topology preservation of the maps and this problem is not tackled by the TS SOM. Tree structured MLP networks like the Perceptron Trees [Utg88] are quite comparable to decision trees as they use an attribute test at each decision node. But unlike in decision trees, now the terminal nodes are not class assignment nodes, but Linear Treshold Units (LTU) which divide the current subspace by a hyperplane. Thus, the terminal nodes require an ....

P. Utgoff. Perceptron trees: A case study in hybrid concept representations. In Proc. of the Nat. Conf. on AI, pages 601--606, St. Paul, MN, 1988.

....used to construct decisions that minimize the distances of the misclassified points from the separating plane. Linear programming approaches [BM92, Ben92, Glo90] find optimal decisions by this criterion in polynomial time. Decision tree methods based on heuristic variants of perceptron algorithms [Utg89, BU92] have worked well in practice, but the algorithms may fail to converge and may not find optimal solutions. The problem of creating a linear function that minimizes the number of points misclassified is NP complete [Hea92] The algorithms CSADT [HKS93] and OC1 [MKSB93, MKS94] use simulated ....

P. E. Utgoff. Perceptron trees: A case study in hybrid concept representations. Connection Science, 1(4):377--391, 1989.

.... there is some evidence that this bias helps in the acquisition of naturally occurring concepts (Spackman, 1988) For example, a successful medical expert system makes use of criteria tables that are essentially m of n concepts (Kingsland, 1985) Our motivation is somewhat similar to that of Utgoff (1988) in developing perceptron trees. In particular, the terms constructed to serve as tests at nodes in the decision tree serve as a representational bias for the learner. However, m of n concepts are less expressive than perceptrons, and provide further constraints on the learning. Furthermore, the ....

.... m of n hypotheses generated by the previous algorithms are used to create new terms that serve as nodes in decision trees (see Table 2) Since the space of m of n concepts is a superset of the space of single attribute discriminations, decision trees constructed in this manner are complete (Utgoff, 1988) in that they can represent any subset of the instance space. Since decision trees also have this property, the intent is not to make decision trees more expressive, but rather to bias decision trees to make polythetic (Fisher, 1987) discriminations. Furthermore, since each node of an m of n ....

Utgoff, P. (1988). Perceptron trees: A case study in hybrid concept representations. Proceedings of the Seventh National Conference on Artificial Intelligence (pp. 601--606). Boston, MA: Morgan Kaufmann.

....explored. Introduction We initially examine a non incremental algorithm that learns binary classification tasks by producing decision trees of linear threshold units (LTU trees) This decision tree bears some similarity to the decision trees produced by ID3 (Quinlan 1983) and Perceptron Trees (Utgoff 1988), yet it seems to promise more generality as each node in our tree implements a separate linear discriminant function while only the leaves of a Perceptron Tree have this generality and the remaining nodes in both the Perceptron Tree and the trees produced by ID3 perform a test on only one ....

Utgoff, P. E. 1988. Perceptron Trees: A Case Study in Hybrid Concept Representation. In AAAI-88 Proceedings of the Seventh National Conference on Artificial Intelligence, 601-6. San Mateo, CA: Morgan Kaufmann.

....is a linear function neuron [326, 188] which can be trained to optimize the sum of distances of the misclassified objects to it, using a convergent procedure for adjusting its coefficients. Perceptron trees, which are decision trees with perceptrons just above the leaf nodes, were discussed in [480]. Decision trees with perceptrons at all internal nodes were described in [482, 438] Mathematical Programming: Linear programming has been used for building adaptive classifiers since late 1960s [216] Given two possibly interesecting sets of points, Duda and Hart [117] proposed a linear ....

....of OC1 s perturbations is monotonically decreasing unlike that of CART LC. 46 LMDT: Another oblique decision tree algorithm, one that uses a very different approach from CARTLC, is the Linear Machine Decision Trees (LMDT) system [483, 48] which is a successor to the Perceptron Tree method [480, 482]. Each internal node in an LMDT tree is a Linear Machine [364] The training algorithm presents examples repeatedly at each node until the linear machine converges. Because convergence cannot be guaranteed, LMDT uses heuristics to determine when the node has stabilized. To make the training stable ....

Paul E. Utgoff. Perceptron trees: A case study in hybrid concept representations. Connection Science, 1(4):377--391, 1989.

....learning. Several algorithms proposed in the literature can be seen as empirical multi strategy learners, but combining different paradigms from RISE s: decision trees, IBL and linear machines [ Brodley, 1993 ] decision trees and rules [ Quinlan, 1987 ] decision trees and perceptrons [ Utgoff, 1989 ] rules and Bayesian classification [ Smyth et al. 1990 ] back propagation and genetic algorithms [ Belew et al. 1992 ] etc. Quinlan [ Quinlan, 1993b ] has successfully combined IBL with trees and other methods, but for the purpose of regression as opposed to classification, performing ....

P. E. Utgoff. Perceptron trees: A case study in hybrid concept representations. Connection Science, 1:377--391, 1989.

....then for some datasets it will be difficult to learn or define areas of expertise. In contrast to randomized methods for creating diversity, top down methods explicitly assign a different classifier for each mutually exclusive subset of the data (Tcheng, Lambert, C Y Lu Rendell, 1989; Utgoff, 1989; Brodley, 1995b) This approach differs from methods that train each base level classifier on a different, randomly selected subset of the data in two ways: 1) in top down methods regions of expertise are defined before learning the base level classifiers rather than afterward and 2) the subsets ....

Utgoff, P. E. (1989). Perceptron trees: A case study in hybrid concept representations. Connection Science, 1, 377-391.

No context found.

Utgoff, P. E. (1989). Perceptron trees: A case study in hybrid concept representations. Connection Science, 1, 377-391.

No context found.

Utgoff, P. E. (1989). Perceptron trees: A case study in hybrid concept representations. Connection Science, 1, 377-391.

No context found.

Utgoff, P. E. (1988). Perceptron trees: A case study in hybrid concept representations. Proceedings of the Seventh National Conference on Artificial Intelligence (pp. 601-606). Saint Paul, MN: Morgan Kaufmann.

....includes univariate decision trees, the heuristic nature of LMDT s search may result in selecting a test from an inappropriate part of the hypothesis space. A solution to this problem would be to determine the appropriate bias dynamically for each test in the tree. The perceptron tree algorithm (Utgoff, 1989) is one example of a system that tries to determine the appropriate representational bias for the instances automatically. Specifically, the algorithm first tries to fit a linear threshold unit(LTU) to the space of instances. If the space is not linearly separable, then the bias of an LTU is ....

Utgoff, P. E. (1989). Perceptron trees: A case study in hybrid concept representations.

No context found.

Utgoff, P. E. (1989). Perceptron trees: A case study in hybrid concept representations. Connection Science, 1(4), 377--391.

No context found.

PAUL E. UTGOFF, Perceptron Trees: A Case Study in Hybrid Concept Representations, Proc. of the 7th Nat. Conf. on AI, St. Paul, Minn., AAAI Press, Menlo Park. Calif., 1988, p601-606. L

No context found.

Utgoff, P., E., (1989) Perceptron Trees: A case study in Hybrid Concept Representations. Connection Science, Volume 1.

First 50 documents

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