Touretzky, D.S. 1986. BoltzCONS: Reconciling connectionism with the recursive nature of stacks and trees, Proceedings Eighth Annual conference of the Cognitive Science Society (Erlbaum, Hillsdale, NJ).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....some idealized brain like network) might realize the types of processes that conventional cognitive science has hypothesized. Connectionists do sometimes explicitly take their models to be theories of implementation. Ballard (1986) even refers to Connectionism as the implementational approach . Touretzky (1986) clearly views his BoltzCONS model this way; he uses Connectionist techniques to implement conventional symbol processing mechanisms such as pushdown stacks and other 37 LISP facilities. Rumelhart McClelland (1986a, p 117) who are convinced that 36. Rumelhardt ....

....the rhetoric fails to keep the implementation algorithm levels distinct. This leads to talk about emergent properties and to the claim that even when they implement Lisp like mechanisms, Connectionist systems can compute things in ways in which Turing machines and von Neumann computers can t (Touretzky, 1986). Such a claim suggests that Touretzky distinguishes different ways of computing not in terms of different algorithms, but in terms of different ways of implementing the same algorithm. While nobody has proprietary rights to terms like ways of computing , this is a misleading way of putting it; ....

Touretzky, D.S. (1986). BoltzCONS: Reconciling connectionism with the recursive nature of stacks and trees, Procedings of the Eighth Annual Conference of the Cognitive Science Society, Amherst, Mass. August, 1986], Hillsdale, N.J: Earlbaum.

....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; Smolensky 1990) The conclusion must be that the PDP Principle concerning representations, 1)a, is consistent with both crucial discreteness and crucial non discreteness of mental representations. In similar vein, consider the second PDP ....

Touretzky, D. S. (1986). BoltzCONS: Reconciling connectionism with the recursive structure of stacks and trees. In Proceedings of the Eighth Annual Conference of the Cogntiive Science Society (pp, 522-530). Hillsdale, NJ: Lawrence Erlbaum.

....of unit x. f = y (we use angle brackets to denote units by indicating their symbolic meaning ) The units thus representing the edges of the f structure DAGs are called e units. The scheme adopted here is a localist version of the recursive representation of LISP S expressions in BoltzCONS [8]. Given a set of features F , a set of atomic feature values V , and set of internal f structure nodes N (which should be regarded as a pool out of which structure representations can be allocated) a total of N . F . N V ) e units would be needed to allow any f structure ....

David S. Touretzky, "BoltzCONS: Reconciling Connectionism with the Recursive Nature of Stacks and Trees". In Proceedings of the 8th Annual Conference of the Cognitive Science Society, Amherst, Mass., August 1986, pp. 522-530.

....structures to be manipulated through a time domain multiplexing technique. It presupposes phased locked loop circuitry whose neuroanatomical reality is highly controversial. Henderson [5] has recently shown that TSVB is sufficient to implement a sophisticated description based parser. Touretzky [15] showed how to emulate a standard lisp engine in a Boltzmann network, thus reintroducing the advantages of von Neumann processing, but also much of its seriality. In the remainder of this paper, I will discuss work in progress on new techniques for mitigating the binding problem in behavior based ....

....model the blocks world by using separate behaviors for each possible set of arguments to the move operator. Previous attempts to solve this problem have usually involved either layering a standard reasoner on top of a behavior based system [2] or building a lisp machine out of neurons or behaviors [15]. An alternative solution is to implement a carefully limited version of binding. Rolepassing is such a technique. It allows role names to be used as simple pointers without implementing centralized routers or forcing the translation of sensory representations into a standard interlingua. The ....

D. S. Touretzky. Boltzcons: Reconciling connectionism with the recursive nature of stacks and trees. Artificial Intelligence, 46(1--2), 1990.

....were carefully chosen so as keep this number close to 28. 49 (A p q) B r C) r q D) D p q A r C B Figure 2.22: A binary tree and the triples which represent it. that triple will be active. However, if other triples have been deleted from memory, some of the units may have been turned off. Touretzky [1990] suggests that a threshold of 75 of units should be used to decide whether a triple is present in memory. DCPS can reliably store around 6 or 7 distinct triples. The probability of error on retrieval increases with the number of triples stored, and with similarity among the stored triples. This ....

Touretzky, D. S. 1986a. Boltzcons: Reconciling connectionism with the recursive nature of stacks and trees. In Proceedings of the Eighth Annual conference of the Cognitive Science Society, Hillsdale N.J. Erlbaum.

....as in(triangle,square) or between(triangle,square,circle) Once propositional representations of simple spatial relations, suitably temporally marked, are computed by local or distributed nodes in a neural network, it remains to do inference with them. Several authors (e.g. Pollack [3] Touretsky [6]) have shown that one can do production system like inference in a neural network, but have not shown any particular advantages of doing it this way. It is therefore not necessary to build inference systems explicitly in terms of neural networks, since one can always re implement standard systems ....

D.S. Touretzky. Boltzcons: Reconciling connectionism with the recursive structure of stacks and trees. Proceedings of the 8th Annual Conference of the Cognitive Science Society, pages 155--64, 1986.

....vector. Once multiple elements can be present, conventional groupings of the elements can be interpreted as larger structures. For example, Touretzky has developed a coarse coded memory system and used it in a production system [20] a primitive lisp data structuring system called BoltzCONS [21], and a combination of the two for simple tree manipulations [22] In his representation, the 15,625 triples of 25 symbols (A Y) are elements to be represented, and using patterns over 2000 bits, small sets of such triples could be reliably represented. Interpreting the set of triples as ....

D. S. Touretzky, BoltzCONS: Reconciling connectionism with the recursive nature of stacks and trees, Proceedings of the 8th Annual Conference of the Cognitive Science Society, Amherst, MA, 1986, 522-530.

....Introduction In the late 1980 s a number of new connectionist models were developed in response to criticisms (e.g. Fodor Pylyshyn, 1988) that connectionism lacked the flexibility and representational adequacy needed for higher level cognitive tasks. Chief among these were coarse coding (Touretzky, 1986), tensor based representation (Smolensky, 1990) reduced representations (Hinton, McClelland Rumelhart, 1986) and RAAM (Pollack, 1990) Compared to earlier systems, they had the advantage of compositionality built more explicitly into their design, and they have shown a great deal of promise in ....

Touretzky, D.S. 1986. BoltzCONS: Reconciling connectionism with the recursive nature of stacks and trees, Proceedings Eighth Annual conference of the Cognitive Science Society (Erlbaum, Hillsdale, NJ).

....research, for making the interpretations [2] Thus, the solution to the first problem requires that neural networks encode schemas, or in general, structured knowledge; that is, it requires addressing the second problem. One approach is to represent such knowledge symbolically in neural networks [3, 4, 5]. The approach works in simple cases but does not generalize well to more complex tasks. Neural networks are not very good at manipulating symbols explicitly. However, they are good at feature extraction, association, constraint satisfaction, pattern classification, and making other fuzzy ....

D. S. Touretzky. BoltzCONS: Reconciling connectionism with the recursive nature of stacks and trees. In Proceedings of 8th Annual Conference of Cognitive Science Society, pages 522-- 530, 1986.

....could develop semantically interpretable higher order elements. And, although the old new connectionism, circa 1980 1985 (cf. Waltz and Pollack (1985) was fairly weak on representational issues, there is currently quite a focus on representation in the new new connectionism (Pollack, 1988; Touretzky, 1986; Touretzky Geva, 1987) One reason that it has taken connectionism several years to rediscover the law of representational adequacy is that there was no body of work in AI involving issues of knowledge representation in neural models. But the real problem which terminated the research viability ....

TOURETZKY, D. S. (1986). BoltzCONS: Reconciling connectionism with the recursive nature of stacks and trees. In Proceedings of the 8th Annual Conference of the Cognitive Science Society. Amherst, MA, 522-530.

....to be manipulated through a time domain multiplexing technique. It presupposes phased locked loop circuitry whose neuroanatomical reality is highly controversial. Henderson (Henderson 1994) has recently shown that TSVB is sufficient to implement a sophisticated descriptionbased parser. Touretzky (Touretzky 1990) showed how to emulate a standard lisp engine in a Boltzmann network, thus reintroducing the advantages of vonNeumann processing, but also much of its seriality. Appendix B: Implementation of naming and dynamic allocation One might reasonably fear that by presupposing dynamic allocation of ....

....active cell deactivates itself. Finally, any cell that notices its name on the left child bus sets its right child register to the name on the active bus. While complicated, this does actually work, and it s infinitely simpler than ephemeral GC. Alternatively, one could use Touretzky s techniques (Touretzky 1990). Ludwig contains a store of 20 cells that can dynamically simulate multiple shift registers in this way. Note that although these simulation techniques have been successfully implemented and tested, they are no longer used on Ludwig for speed reasons. Ludwig is implemented a register transfer ....

D. S. Touretzky. Boltzcons: Reconciling connectionism with the recursive nature of stacks and trees. Artificial Intelligence, 46(1--2), 1990.

....mechanisms with abstract representations of events. Let us remark as a tendency in accordance with this approach the growing emphasis given recently to hybrid symbolic connectionist models of cognition (e.g. Hall Romaniuk, 1990; Lange et al. 1989; Lee et al. 1989; Rose Bellew, 1989; Touretzky, 1986). 1.2. Associationism and Weighting Mechanisms in Machine Learning Weights, or numeric values attached to hypotheses generated symbolically, are used in inductive machine learning in order to reflect information about the number of examples correctly or incorrectly classified by a learning ....

Touretzky, D. (1986): BoltzCONS: Reconciling connectionism with the recursive nature of stacks and trees. In Proceedings of the Eighth Annual Conference of the Cognitive Science Society. Hillsdale, NJ: Erlbaum.

No context found.

Touretzky, D.S. 1986. BoltzCONS: Reconciling connectionism with the recursive nature of stacks and trees, Proceedings Eighth Annual conference of the Cognitive Science Society (Erlbaum, Hillsdale, NJ).

No context found.

Touretzky, David S. (1986), "BoltzCONS: Reconciling Connectionism with the Recursive Nature of Stacks and Trees". In Proceedings of the 8th Annual Conference of the Cognitive Science Society, Amherst, Mass., pp. 522-530.

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