J. A. Feldman and D. H. Ballard, Connectionist models and their properties, Cognitive Science, 6 (1982), pp. 205--254.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....most interesting explored transitions probabilistically bias future search, preventing ants to waste resources in not promising regions of the search space. Neural networks. Ant colonies, being composed of numerous concurrently and locally interacting units, can be seen as connectionist systems [42], the most famous examples of which are neural networks [3, 61, 85] From a structural point of view, the parallel between the ACO meta heuristic and a generic neural network is obtained by putting each state i visited by ants in correspondence with a neuron i, and the problemspeci c neighborhood ....

J. A. Feldman and D. H. Ballard. Connectionist models and their properties. Cognitive Science, 6:205-254, 1982. 32

....descendant of associative memory ideas. Memory based reasoning holds considerable promise, both for cognitive modeling and for applications. In this model, rote memories of episodes play the central role, and schemas are viewed as epiphenomenal. This model is described in considerable detail in [351 and will not be explained here; however, as I have prepared this paper, it has served as the background against which .I have critically examined both connectionist and more traditional AI paradigms. 2 Connectionist and Heuristic Search Models For most of its history, the heuristic search, ....

....[9] Meanwhile, it seems highly implausible that anything resembling heuristic search is used much below the level of consciousness; certainly no one would believe that a neuron executes heuristic search. The small amount of evidence marshalled to support the hypothesis of subconscious search [15] could be explained in many other ways. Such models as Marcus deterministic parser [29] have attempted to move away from heuristic search, yet were cast largely in heuristic search terms 1One problem that Marcus parser was attempting to solve was the ntismatch between psychological data and ....

[Article contains additional citation context not shown here]

Feldman, J.A., Ballard, D.H., "Connectionist Models and Their Properties," Cognitive Science 6 (3), 205-254, 1982.

....of coupled subnets. Thus this approach is an emancipation from a homunculus who controls the feature extraction processes and finally selects the particular class. One way to deal with the maximum selection from a set of inputs within a connectionist framework are winner take all (WTA) networks ([3]) The operation of these networks is a mode of contrast adjustment and pattern normalization where only the unit with the highest activation fires and all other units in the network are inhibited after some setting time. An example of a common competitive architecture to select the maximum or ....

....where only the unit with the highest activation fires and all other units in the network are inhibited after some setting time. An example of a common competitive architecture to select the maximum or minimum from a set of data is MAXNET ( 10] Other techniques to pick a maximum can be found in ([3], 5] 9] 11] In [12] the maximum selection is generalized to k winners take all networks which identify the largest k of n real numbers. Robust WTA architectures with evidential response are considered in ( 7] 8] The overall architecture of the self organizing classification network ....

J.A. Feldmann and D.H. Ballard. Connectionist models and their properties. Cognitive Science, 6:205--254, 1982.

....dealing with com monsense reasoning in general (see [16] 3 which is used in this work to carry out the intensional approach with a two level representation. We adopt this architec ture for the following reasons (as will be demonstrated later on) aFor similar connectioinist models, see [1] [5], 6] etc. It provides a computationally efficient (massively parallel) mechanism. It provides a way for implementing an intensional semantical (vs. exten sional syntactical) approach to the inheritance problem. The inheritance problem is reduced to the more general problem of rule ap ....

....activated if a concept similar to them is activated in CD. Finally in the bottom up phase, fully or partially activated CD (feature) representations will go back up to activate the corresponding nodes in CL. The result can be read off from CL. s 9Each node in the system has one or more sites (cf. [5]) each of which computes the weighted sum (or any other similar functions whenever needed) of the inputs. The maximum of the values computed by all the sites is taken to be the activation value of the node. 17 Notice the massive parallelism in the above specified architecture: activations are ....

J. Feldman and D. Ballard, Connectionist models and their properties, Cognitive Science, pp.205-254, July 1982

....block of the human nervous system, the neuron, is quite slow. Its computational speed is a few milliseconds and must account for the complex behavior carried out in a few hundred milliseconds [Posner, 1978] This means that entire complex behaviors are carried out in less than a hundred time steps [Feldman and Ballard, 1982]. Within the field of Artificial Intelligence many attemps were made to model such complex behaviors. But current models need millions of time steps. Moreover, many of the problems that humans can solve almost instanteneously as if it were a reflex become intractable if modelled in a conventional ....

....system and connectionist techniques were used to learn a heuristic from examples. An outlook to current research problems completes the article. 2 Connectionist Systems Connectionist systems aim at modeling aspects of the animal and human nervous system on an abstract computational level (cf. [Feldman and Ballard, 1982; Rumelhart et al. 1986] The central concept in a connectionist system is the individual unit . It models the functionality of a neuron or a group of neurons. Following [Feldman and Ballard, 1982] each unit is characterized by p the potential , which is a real number, v the value, which is an ....

[Article contains additional citation context not shown here]

J. A. Feldman and D. H. Ballard. Connectionist models and their properties. Cognitive Science, 6(3):205--254, 1982.

.... There is evidence to suggest that information processing takes place in the dendritic trees of biological neurons (see [3] for a review) Local thresholding of the summed input from clusters of excitatory synapses within the dendritic tree could account for the multiplicative responses of cells [4, 5]. A single such unit is thus as powerful as a multilayer network of linear threshold units, and the outputs of many nonlinear units can be combined together to act as a larger, virtual , nonlinear unit [5] 1.1 Sigma Pi Units The principal model of a nonlinear neuron is the sigma pi unit ....

J. A. Feldman and D. H. Ballard. Connectionist models and their properties. Cognitive Science, 6:205--54, 1982.

....a discrimination measure. These properties justify a biologically inspired fault tolerant extension of the network using di erentiating neurons. 1 Introduction One way to deal with the maximum selection from a set of inputs within a connectionist framework are winner take all (WTA) networks ([4]) The operation of these networks is a mode of contrast enhancement and pattern normalization where only the unit with the highest activation res and all other units in the network are inhibited after some setting time. References to common competitive architectures to select the maximum or ....

J.A. Feldmann and D.H. Ballard. Connectionist models and their properties. Cognitive Science, 6:205-254, 1982.

....led to the postulation of independent feature maps with only weak cross bindings. Another, closely related way to view this phenomenon is in terms of the dissociation of feature type and feature location. Then the cross talk results from the failure to bind two properties of the same feature. Feldman and Ballard (1982) view focused attention as the mechanism typically use to fix the location of a visual object and thereby allow independent feature properties associated with that object to be bound together. The importance of focused attention for feature conjoining is also stressed by Treisman Gelade (1980) ....

Feldman, J.A. & Ballard, D.H. (1982). Connectionist models and their properties. Cognitive Science, 6, 205-254.

....to the fact that neurons are comparatively slow computing devices, there seems to be only the one conclusion This work is supported by the Deutsche Forschungsgemeinschaft (DFG) within project MPS under grant no. HO 1294 3 2. that massive parallel computations must be performed (see e.g. [7]) L. Shastri and V. Ajjanagadde have developed a massively parallel, connectionist computational model for a limited class of inference problems called Shruti [13, 14] The number of computing elements is bound by the size of the knowledge base and queries are answered with respect to that ....

J. A. Feldman and D. H. Ballard. Connectionist models and their properties. Cognitive Science, 6(3):205--254, 1982.

....explicitly concerned with the problem of internal representation (p 121) Correspondingly, the specification of what the states of a network represent is an essential part of a Connectionist model. Consider, for example, the well known Connectionist account of the bistability of the Necker cube (Feldman and Ballard, 1982). Simple units representing the visual features of the two alternatives are arranged in competing coalitions, with inhibitory. links between rival features and positive links within each coalition. The result is a network that has two dominant stable states (See Figure 1) Notice that, in ....

....brains. These may be seen as favoring the Connectionist alternative. Below we will sketch a number of these before discussing the general problems which they appear to raise. Rapidity of cognitive processes in relation to neural speeds: the hundred step constraint. It has been observed (e.g. Feldman Ballard, 1982) that the time required to execute computer instructions is in the order of nanoseconds, whereas neurons take tens of milliseconds to fire. Consequently, in the time it takes people to carry out many of the tasks at which they are fluent (like recognizing a word or a picture, either of which may ....

Feldman, J.A. & Ballard, D.H. (1982). Connectionist models and their properties. Cognitive Science, 6, 205-254.

....assigns biases to sites, implemented through a special BiasUnit whose site is connected to every other site in the net with a constant activation of 1. 2.4Structure Modi ficati on 2.4. 1Bi ologi cal Moti vati on Connectionists network researchers look to Neurobiology for inspiration (Feldman Ballard, 1982), Reek e Edelman, 1988) However, where much has been gained from the general concept of network s of simple units, little progress has been made towards a general solution for the optimal structure of such a net. Recently, there has been growing interest in dynamic adjustment of the ....

Feldman, J.A., & Ballard, D.H. 1982. Connectionist models and their properties.

.... as an approximative low order polynomial function [12] Other similar models have been proposed by different researchers, and the clusterons are also known as functional link units [15,20] and oe neural units [3] Various applications involve clusteron based neural networks in engineering [4], ranging from the field of automatic control (see for instance [9] to the area of communications (see e.g. 10] Statistics estimation by non parametric functions is a widely addressed problem in the Neural Networks literature [2,14,1,13,16 19,11] Particularly, over the recent years much ....

J.A. Feldman and D.H. Ballard, Connectionist models and their properties, Cognitive Science, Vol. 6, pp. 205 -- 254, 1982

....a threshold circuit with constant depth. There has been a considerable amount of renewed interest in models for the brain and for learning, and many of the recently proposed models are again essentially constant depth Threshold Circuits. Examples of these models include the Connectionist Models [5] and the Boltzmann Machine [1, 10] Recently, Parberry and Schnitger proved that Boltzmann Machines can be simulated by constant depth Threshold Circuits [16] This paper is a further theoretical investigation of bounded depth Threshold Circuits. In particular, we consider the following ....

J. A. Feldman and D. H. Ballard, Connectionist models and their properties, Cognitive Science, 6 (1982), pp. 205--254.

.... plays a crucial role in the important class of models with continuous units that converge to discrete representations (e.g. Anderson, Silverstein, Ritz and Jones, 1977; Rumelhart, Smolensky, McClelland and Hinton, 1986) including models with winner take all sub networks (e.g. Grossberg, 1976; Feldman and Ballard, 1982; Rumelhart and Zipser, 1985; Mozer, 1991. And of course discreteness of representations is also a central property of a number of connectionist techniques for embedding symbolic structures as patterns of activity (e.g. Touretzky, 1986; Touretzky and Hinton, 1988; Dolan, 1989; Pollack, 1990; ....

Feldman, J. A., & Ballard, D. H. (1982). Connectionist models and their properties. Cognitive Science, 6, 205-254.

.... Fax: 49 351 463 8342 sh inf.tu dresden.de Keywords: Logic Programming, Recurrent Connectionist Networks 1 1 Motivation Connectionist systems exhibit many desirable properties of intelligent systems like, for example, being massively parallel, context sensitive, adaptable and robust (see e.g. [7]) It is strongly believed that intelligent systems must also be able to represent and reason about structured objects and structure sensitive processes (see e.g. 8, 23] Unfortunately, we are unaware of any connectionist system which can handle structured objects and structure sensitive ....

J. A. Feldman and D. H. Ballard. Connectionist models and their properties. Cognitive Science, 6(3):205-254, 1982.

....for dealing with commonsense reasoning in general [16] 3 which is used in this work to carry out the intensional approach with a two level representation. We adopt this architecture for the following reasons (as will be demonstrated later on) 3 For similar connectioinist models, see [1] [5], 6] etc. 5 ffl It provides a computationally efficient (massively parallel) mechanism. ffl It provides a way for implementing an intensional semantical (vs. extensional syntactical) approach to the inheritance problem. ffl The inheritance problem is reduced to the more general problem of ....

.... subsymbolic level) and thus forming a two level architecture, one can solve problems by utilizing the synergy between the two components ( 16] 17] and [22] CONSYDERR handles all inheritance relations with two primitives: rules 9 Each node in the system has one or more sites (cf. [5]) each of which computes the weightedsum (or any other similar functions whenever needed) of the inputs. The maximum of the values computed by all the sites is taken to be the activation value of the node. 16 (e.g. elephant Gamma color gray) implemented with links in CL and CD, and ....

J. Feldman and D. Ballard, Connectionist models and their properties, Cognitive Science, pp.205-254, July 1982

....interpretations of nodes or links are possible. Often, specific knowledge of the task is built into a unified neural architecture. Much early research on unified neural architectures can be traced back to work by Feldman and Ballard, who provided a general framework of structured connectionism [16]. This framework was extended in many di#erent directions including, for instance, parsing [14] explanation [12] and logic reasoning [30, 40, 70 72] Recent work along these lines focuses also on the so called NTL, Neural Theory of Language, which attempts to bridge the large gap between ....

.... [77] Localist connectionist architectures contain one distinct node for representing each concept [42, 71, 67, 3, 58, 31, 66] Distributed neural architectures comprise a set of non exclusive, overlapping nodes for representing each concept [60, 50, 27] The work of researchers such as Feldman [16, 17], Ajjanagadde and Shastri [67] Sun [72] and Smolensky [69] has demonstrated why localist connectionist networks are suitable for implementing symbolic processes usually associated with higher cognitive functions. On the other hand, radical connectionism [13] is a distributed neural approach to ....

J. A. Feldman and D. H. Ballard. Connectionist models and their properties. Cognitive Science, 6:205--254, 1982.

No context found.

J. A. Feldman and D. H. Ballard, Connectionist models and their properties, Cognitive Science, 6 (1982), pp. 205--254.

No context found.

J. A. Feldman and D. H. Ballard. Connectionist models and their properties. Cognitive Science, 6(3):205--254, 1982.

No context found.

J. Feldman, D. Ballard. Connectionist models and their properties, Cognitive Science 6, 205 - 254. (1982).

No context found.

J.A. Feldman and D.H. Ballard. Connectionist models and their properties. Cognitive Science, 6(3), 1982.

No context found.

J.A. Feldman. Connectionist models and their properties. Cognitive Science, 6:205-- 254, 1982.

No context found.

J. A. Feldman and D. H. Ballard. Connectionist models and their properties. Cognitive Science, 6:205--54, 1982.

No context found.

Feldman JA, Ballard DH (1982) Connectionist models and their properties. Cogn Sci 6: 205--254

No context found.

J. Feldman and D. Ballard, "Connectionist Models and Their Properties", Cognitive Science, Vol. 6, No. 3, 1982,

First 50 documents  Next 50

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