Feldman, J.A. (1986). Neural representation of conceptual knowledge. Report TR189. Department of Computer Science, University of Rochester.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....an entire population of units does the encoding (the so called distributed representation networks) is considered to be important by many people working on Connectionist models. Although Connectionists debate the relative merits of localist (or compact ) versus distributed representations (e.g. Feldman, 1986), the distinction will usually be of little consequence for our purposes, for reasons that we give later. For simplicity, when we wish to refer indifferently to either single unit codes or aggregate distributed codes, we shall refer to the nodes in a network. When the distinction is relevant to ....

Feldman, J.A. (1986). Neural representation of conceptual knowledge. Report TR189. Department of Computer Science, University of Rochester.

....natural language understanding or problem solving systems. This work describes an attempt to develop computational models of causal semantics based on a synthesis of results in sensory motor control #Kandel et al. 1991; Latash 1993; Sternberg 1978#, insights from structured connectionist systems #Feldman 1989# and linguistic research in CognitiveSemantics. Furthermore, we exploit the fact that the deep semantics of the causal narratives are dynamic and arise from a continuous interaction between input and memory. This enables the high degree of context sensitivity required since changing input context ....

Feldman, J.A. #1989# Neural Representation Of Conceptual Knowledge, Nadel, L.

....the time complexity of computation, but in terms of ease of implementing it in a connectionist fashion with a set of simple, autonomous, locally connected nodes. We want (1) all computation to be local, 2) only simple messages to be passed around, and (3) no extra nodes to be added (see Feldman [7] for similar points) With these three criteria in mind, again only the last model can be selected (details are omitted) In Figure 7, suppose A is externally activated, then because of the microfeatures shared with A, B will later be activated. The activation of A will go top down to its ....

J. Feldman, Neural Representation of Conceptual Knowledge, Technical Report 189,Department of Computer Science, University of Rochester, 1986

....such as John corresponds to a focal node of the representation of the entity John . Information about the various features of John and the roles he fills in various events is encoded by linking the focal node to appropriate nodes distributed throughout the network (see Shastri Feldman, 1986; Feldman, 1989). Encoding of Predicates: Predicate Clusters as Convergence Zones Consider the encoding of the binary predicate love with two roles: lover and lovee. This predicate is encoded by a cluster of nodes consisting of two role nodes depicted as circular nodes and labeled lover and lovee; an enabler ....

Feldman, J. A. (1989) Neural Representation of Conceptual Knowledge. In L. Nadel, L.A. Cooper, P. Culicover, & R.M. Harnish (Eds.), Neural Connections, Mental Computation (pp. 68--103). Cambridge, MA: MIT Press.

....natural language understanding or problem solving systems. This work describes an attempt to develop computational models of causal semantics based on a synthesis of results in sensory motor control (Kandel et al. 1991; Latash 1993; Sternberg 1978) insights from structured connectionist systems (Feldman 1989) and linguistic research in Cognitive Semantics. Furthermore, we exploit the fact that the deep semantics of the causal narratives are dynamic and arise from a continuous interaction between input and memory. This enables the high degree of context sensitivity required since changing input context ....

Feldman, J.A. (1989) Neural Representation Of Conceptual Knowledge, Nadel, L.

....Regier s model, in particular, the notion that space serves as a fundamental conceptual structuring device in language (p. 19) The brand of connectionism Regier chose to use to model closed class lexeme acquisition is also indebted to Jerome Feldman s so called structured connectionist models (Feldman, 1989; Feldman, Fanty, and Goddard, 1988) These are essentially networks whose nodes have symbolic interpretations (e.g. dog might be a node in such a network) and whose architecture specifically reflects various cognitive structures. Regier constrains his network by explicitly incorporating a ....

Feldman, J. (1989). Neural representation of conceptual knowledge. In Neural connections, mental computation, L. Nadel, P. Culicover, R. M. Harnish (eds.). Cambridge, MA: MIT Press.

.... Baillet, and Brown 1984) The evidence for the automatic occurrence of elaborative or predictive inferences however, is mixed (see e.g. Kintsch 1988; Potts, Keenan, and Golding 1988) 2 For an overview of the structured connectionist approach and its merits see (Feldman and Ballard, 1982; Feldman, 1989; Shastri, 1995) 3 http: www.icsi.berkeley.edu shastri shruti index.html 1 ffl inference involving relational knowledge corresponds to a transient propagation of rhythmic activity across such focal clusters, ffl a binding between a conceptual role and an entity filling that role in a given ....

Feldman, J. A. (1989) Neural Representation of Conceptual Knowledge. In Neural Connections, Mental Computation ed. L. Nadel, L.A. Cooper, P. Culicover, and R.M. Harnish. MIT Press.

.... kinds of representations being proposed and studied in the neural network area: local representation (LR) also called grandmother cell representation, and distributed representation (DR) Originally, the LR is defined as the one in which each entity is represented by a single unit (Barlow, 1972; Feldman, 1986), and the DR is defined as the one in which each entity is represented by a pattern of activity distributed over many units, and each unit is involved in representing many different entities (Hinton, 1981; Hinton, McClelland and Rumelhart, 1986; Anderson and Rosenfeld, 1988) However, there ....

....only concentrate on the element to unit (for LR, it is the pattern to unit) level mainly. With these correspondences, we will be able to discuss various representations more specifically. 4. CORRESPONDENCE CLASSIFICATION 4.1. Local Representation (LR) The original LR defined by (Barlow, 1972; Feldman, 1986) is the pattern to unit one to one correspondence defined above. LR is best characterized by the pattern to unit correspondence because in LR the pattern is the smallest constituent of the entity. A pattern is not further divided into more elements (each pattern contains only one element) ....

Feldman, J. A. (1986). Neural representation of conceptual knowledge. Technical Report TR 189, Department of Computer Science, University of Rochester, Rochester, NY.

....the time complexity of computation, but in terms of ease of implementing it in a connectionist fashion with a set of simple, autonomous, locally connected nodes. We want (1) all computation to be local, 2) only simple messages to be passed around, and (3) no extra nodes to be added (see Feldman [7] for similar points) With these three criteria in mind, again only the last model can be selected (details are omitted) In Figure 7, suppose A is externally activated, then because of the microfeatures shared with A, B will later be activated. The activation of A will go top down to its ....

J. Feldman, Neural Representation of Conceptual Knowledge, Technical Report 189,Department of Computer Science, University of Rochester, 1986

....funds, and subcontracts from Cognitive Technologies Inc. under ONR grant N00014 95 C 0182 and ARI contract DASW01 97C 0038. Work on the CM 5 version of Shruti was supported by NSF Infrastructure Grant CDA 8722788. 2 For an overview of the structured connectionist approach and its merits see [46][166] Neural Computing Surveys 2, 1 40, 1999, http: www.icsi.berkeley.edu jagota NCS 21 shruti suggested that the encoding of relational information (frames, predicates, etc. is mediated by neural circuits composed of focal clusters and the dynamic representation and communication of ....

J.A. Feldman. Neural representation of conceptual knowledge. In P. Culicover & R.M. Harnish L. Nadel, L.A. Cooper, editor, Neural Connections, Mental Computation, volume 1, pages 68--103. MIT Press, Cambridge, MA, 1989.

....of the proposed architecture, for it suggests that the representation of certain types of algebraic rules can be grounded in basic notions of spatial and temporal locations and coincidence. We believe that this work is a simple demonstration of the power of melding structured connectionist models [6] with the notion of temporal synchrony variable binding, and learning. Other examples may be found in [13, 14, 7] Acknowledgments This work was partially funded by ONR grants N00014 93 1 1149 and NSF grant SBR9720398. Thanks to Gary Marcus, Jerry Feldman and other members of the NTL group for ....

J.A. Feldman. Neural representation of conceptual knowledge. In L. Nadel, L.A. Cooper, P. Culicover, and R.M. Harnish, editors, Neural Connections, Mental Computation. MIT Press, Cambridge, MA, 1989.

....increases the learning capacity of NN and permit to apply them to large scale problems [1] This is highly confirmed by the experience carried out by [25] This latter shows that a random pruning of connections before any learning improves significantly the network s performance. Note that [26] and [27] argue that a complex behaviour requires bringing together several different kinds of knowledge and processing, which is not possible without structure (modularity) Usually the MNN implementation are based on the divide and conquer method. This method consists first in breaking down a ....

J. Feldman, "Neural representation of conceptual knowledge," in Neural connections, mental computation (Nadel and al., eds.), (Cambridge, MA), MIT Press, 1989.

....to many advantages compared to the use of single Neural Networks. For instance, to constrain the network connectivity (i.e. to introduce local computation in the network) increases the learning capacity of NN and permits to apply them to large scale problems (Happel and Murre 1994) Note that (Feldman 1989) and (Simon 1981) argue that a complex behaviour requires bringing together several different kinds of knowledge and processing, which is not possible without structure (i.e. modularity) As (Mure et al. 1992) Jacobs et al. 1991a) Bottou and Galliinari 1991) highlight, modularity is a way to ....

Feldman, J. (1989). Neural representation of conceptual knowledge. In Nadel and al. (Eds.). `Neural connections, mental computation'. MIT Press. Cambridge, MA.

....language, generating language, memory storage, memory retrieval, and reasoning. Complex behavior requires bringing together several different kinds of knowledge sources and processes, something that cannot be done in a single pattern transformation. Such behavior requires structured architectures (Feldman, 1989; Minsky, 1985; Simon, 1969) and combinations of different types of networks. 2. The required network size, the number of training examples, and the training time become intractable as the size of the problem grows (Elman, 1991; Harris and Elman, 1989; St. John and McClelland, 1990; St. John, ....

Feldman, J. A. 1989. Neural representation of conceptual knowledge. In: Nadel, L., Cooper, L. A., Culicover, P., and Harnish, R. M. (eds.), Neural connections, mental computation, MIT Press, Cambridge, Massachusetts, pp. 68--103.

....a discussion about this issue see (Ronco and Gawthrop, 1995) Modularity is a natural way to ease the learning of complex behaviour. Mountcastle (Mountcastle, 1978) and many others (Karmiloff Smith, 1994; Houk and Wise, 1995; Carpenter, 1984) argue that modularity is a brain organising principle. (Feldman, 1989) and (Simon, 1981) highlight the fact that a complex behaviour requires the bringing together of several different kinds of knowledge and processing, which is not possible without structure (i.e. modularity) Moreover, and according to Marr such a modularity can enable a learning economy. In case ....

Feldman, J.A. (1989). Neural representation of conceptual knowledge. In: Neural connections, mental computation (Nadel and al., Eds.). MIT Press. Cambridge, MA.

....NN (see for further details (Ronco and Gawthrop, 1995) Modularity is a natural way to ease the learning of complex behaviours. Mountcastle (Mountcastle, 1978) and many others (Karmiloff Smith, 1994; Houk and Wise, 1995; Carpenter, 1984) argue that modularity is a brain organising principle. (Feldman, 1989) and (Simon, 1981) highlight the fact that a complex behaviour requires the bringing together of several different kinds of knowledge and processing, which is not possible without structure (i.e. modularity) Moreover, and according to Marr such a modularity can enable a learning economy. In case ....

Feldman, J.A. (1989). Neural representation of conceptual knowledge. In: Neural connections, mental computation (Nadel and al., Eds.). MIT Press. Cambridge, MA.

....such as parsing language, generating language, memory storage, memory retrieval, and reasoning, which cannot be performed in a single pattern transformation. Complex behavior requires bringing together several different kinds of knowledge and processing, which is not possible without structure (Feldman, 1989; Simon, 1981) 2. The required network size, the number of training examples and the training time becomes intractable as the size of the problem grows, especially in sequential processing (Harris and Elman, 1989; Servan Schreiber et al. 1989) 3. There is no way to evaluate what the entire ....

Feldman, J. A. (1989). Neural representation of conceptual knowledge. In Nadel, L., Cooper, L. A., Culicover, P., and Harnish, R. M., editors, Neural Connections, Mental Computation.

....sparse patterns. The same goes for sparse real valued patterns (ones in which large values are rare) and non negative real valued patterns. Finding a scheme that worked with sparse patterns would be interesting because it seems that the human brain probably uses sparse distributed representations [Feldman 1986]. There may be simple modifications of current binding and superposition operations which will work, or there may be other operations which could be used instead. Different binding operations are another unexplored region. In all of the schemes described, the binding operation uses very ordered ....

Feldman, J. A. (1986). Neural representation of conceptual knowledge. Technical report TR189, Department of Computer Science, University of Rochester, Rochester NY.

....5 love lover lovee Tom Susan John Mary :love :love :love 0 Mary Tom lover lovee time 6 5 4 3 2 1 :love Figure 3: a) The structure of a predicate cluster. b) The rhythmic pattern of activation representing the dynamic bindings love(Mary,Tom) throughout the network (see (Feldman, 1989; Shastri, 1991) Encoding of predicates using focal clusters: Consider the encoding of the binary predicate love with two roles: lover and lovee. This predicate is encoded by a cluster of nodes consisting of two role nodes depicted as circular nodes and labeled lover and lovee; an enabler node ....

Feldman, J. A. (1989) Neural Representation of Conceptual Knowledge. In L. Nadel, L.A. Cooper, P. Culicover, & R.M. Harnish (Eds.), Neural Connections, Mental Computation (pp. 68--103). Cambridge, MA: MIT Press.

....in the memory storage, retrieval, and indexing process. Our observations suggest the use of process semantics #Nielsen et al. 1981# to characterize the deep semantics of causal narratives. Of special interest to us is the possibility of modeling such processes in structured connectionist models #Feldman 1989#. This work attempts to develop computational models based on a synthesis from process theory, insights from structured connectionist systems and from linguistic researchon Cognitive Semantics to model the commonsense understanding of causal narratives. We #rst illustrate these observations by ....

Feldman, Jerome A. 1989. Neural representation of conceptual knowledge. In Neural Connections, Mental Computation, ed. by Lynn Nadel et al., 68#103. Cambridge, Mass.: MIT Press.

No context found.

Feldman, J.A. (1989). Neural representation of conceptual knowledge. In L. Nadel, L.A. Cooper, P. Cullicover, and R.M. Harnish (eds.), Neural Connections, Mental Computation. Cambridge, MA: MIT Press, 68-103.

No context found.

Jerome Feldman, "Neural Representation of Conceptual Knowledge," Technical Report 189, Department of Computer Science, University of Rochester, 1986.

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