L. Shastri. Semantic Networks: An Evidential Formalization and Its Connectionist Realization. Pitman, London, 1988.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....such topologies shown in Figure 1. This work describes a simulation study in order to verify timing hypotheses proposed earlier [22] We hope ultimately our models will help establish a biological grounding for structured connectionist models capable of cognitive functions like reasoning (such as [14, 47, 46, 61, 16, 63, 48]) Our work is consistent with previous work on recruitment learning [47, 49, 14, 9, 16] We augment the neuroidal model to continuous time by using the spike response model (SRM) of Gerstner [18] Complementing other studies on recruitment which mainly provided analytical calculations and ....

.... hypotheses proposed earlier [22] We hope ultimately our models will help establish a biological grounding for structured connectionist models capable of cognitive functions like reasoning (such as [14, 47, 46, 61, 16, 63, 48] Our work is consistent with previous work on recruitment learning [47, 49, 14, 9, 16]. We augment the neuroidal model to continuous time by using the spike response model (SRM) of Gerstner [18] Complementing other studies on recruitment which mainly provided analytical calculations and statistical simulations [61, 17, 49, 16] our work extends by implementing an actual simulator ....

Lokendra Shastri. Semantic networks: An evidential formalization and its connectionist realization. Research Notes in Artificial Intelligence. Morgan Kaufmann Publishers, Inc., San Meteo, California, 1988.

....such topologies shown in Figure 1. This work describes a simulation study in order to verify timing hypotheses proposed earlier [22] We hope ultimately our models will help establish a biological grounding for structured connectionist models capable of cognitive functions like reasoning (such as [14, 47, 46, 61, 16, 63, 48]) Our work is consistent with previous work on recruitment learning [47, 49, 14, 9, 16] We augment the neuroidal model to continuous time by using the spike response model (SRM) of Gerstner [18] Complementing other studies on recruitment which mainly provided analytical calculations and ....

.... hypotheses proposed earlier [22] We hope ultimately our models will help establish a biological grounding for structured connectionist models capable of cognitive functions like reasoning (such as [14, 47, 46, 61, 16, 63, 48] Our work is consistent with previous work on recruitment learning [47, 49, 14, 9, 16]. We augment the neuroidal model to continuous time by using the spike response model (SRM) of Gerstner [18] Complementing other studies on recruitment which mainly provided analytical calculations and statistical simulations [61, 17, 49, 16] our work extends by implementing an actual simulator ....

Lokendra Shastri. Semantic networks: An evidential formalization and its connectionist realization. Research Notes in Artificial Intelligence. Morgan Kaufmann Publishers, Inc., San Meteo, California, 1988.

.... several techniques for knowledge representation and originally applied as a representational language for knowledge acquisition in many aspects of artificial intelligence (AI) 18] The fundamentals of CGs are based on the existential graphs of Charles Sanders Peirce [19] and the semantic networks [20]. CGs combine the expressive power of natural languages with the formality of symbolic logic. With direct mapping to and from natural languages, CGs can serve as an intermediate interpretation language between computer oriented specifications and natural languages. With their graphical notations ....

Shastri L., "Semantic Networks: An Evidential Formalization and Its Connectionist Realization", Research notes in artificial intelligence, Pitman, London, 1988.

....The results of the analysis appear in Figure 5 where no class has two category of superclasses. Hence, RULE 2 is satisfied. Example 2: A class person has two subclasses Quaker and Republican. These classes in turn have a joint subclass Republican Quaker. This example is known as Nixon Diamond [31, 32] (Figure 6) By CASE 1.2 the primary superclass for Republican Quaker is Quaker. There are three contexts in this example. A personal context containing the class person, a religious context containing Quaker and Republican Quaker, and a political context containing Republican. Thus, we define ....

L. Shastri. Semantic Networks: an Evidential Formalization and its Connectionist Realization. Morgan Kaufmann Publishers, San Mateo, CA, 1988.

.... compared to the more or less heuristic models of most other hybrid systems (for our earlier, heuristic attempts towards a hybrid system, see [46 48] Secondly, although some hybrid models provide theoretical justifications for the computations (see e.g. Shastri s system for evidential reasoning [49]) they may require fairly complicated and heterogeneous computing elements and control regimes, whereas the neural network model behind our Bayesian neural system is structurally very simple and uniform, and confirms to an already existing family of neural architectures, the Boltzmann machines. ....

L. Shastri. Semantic Networks: An Evidential Formalization and Its Connectionist Realization. Pitman, London, 1988.

....and will not aid in the resource discovery process. Alternately, the UoD division will be too coarse if every node associates itself with the same global concept. How nodes define themselves in terms of the global concepts is based upon the semantic network notion of weighted relationships links [30, 31]. Each node, independent of all other nodes, will form a one way connection with each global concept. Initially weights will be based on a subjective estimation (by an administrator) of how well a particular node fits within a global concept (see Figure 3) To aid in this process, a defining ....

L. Shastri L, "Semantic Networks: An Evidential Formalization and its Connectionist Realization, Pitman, Morgan Kaufman, 1988.

....proposal consists in advocating the use of Fuzzy Logic as a KR tool, while we suggest to combine DS theory with a KR tool. Other systems have recently been developed in which both the need of representing knowledge and that of representing uncertainty have been taken into consideration (e.g. Shastri, 1988; Yen and Bonissone, 1990; Heinsohn, 1991) Differently from our proposal, these systems only consider a particular KR system: they do not address the general problem of integration. Moreover, they do not consider the issue of isolating formulae from propositions: by contrast, in our approach ....

Shastri, L. (1988) Semantic Networks: An Evidential Formalization and its Connectionist Realization, London: Pitman.

....but feature structures (f structures, for short) i.e. recursively embedded matrices of feature value pairs, much in the spirit of frame like data structures traditionally found in AI. The feasibility and usefulness of such structures in connectionism has been demonstrated in a number of models [2, 5, 9]. On developing our approach, we aimed at fully general representation and processing of arbitrarily embedded structures. The connectionist approach to unification described in this paper was implemented and tested using an interactive network simulator based on LOOPS running on Interlisp ....

Lokenda Shastri, Semantic Networks: An Evidential Formalization and its Connectionist Realization. San Mateo, Calif.: Morgan Kaufmann, 1988.

....3 Sparse Connectionist Networks Connectionist networks consist of simple processing units characterized by an output activation and connected by weighted links. Networks with sparse connectivity are more neurally plausible, and are used for a wide range of highly structured cognitive models (e.g. [6], 7] Sparse networks also result from the application of weight decay techniques (like [8] to densely connected multi layer backpropagation networks. In most connectionist networks, evaluating a unit usually consists of computing the dot product of the input weights and activations, followed ....

L. Shastri. Semantic Networks: Evidential Formalization and its Connectionist Implementation. Morgan Kaufmann, 1988.

....properties from more abstract models higher up in the conceptual hierarchy. Unlike traditional semantic network representations, our approach to the problem of model indexing is based on a special formalism for integrating evidential reasoning into semantic networks [Shastri and Feldman 1985, Shastri 1988b] Within Shastri s evidential formulation, finding solutions to the recognition problem may be viewed as decision making under uncertainty that requires choosing the most likely alternative from among a set of mutually exclusive alternatives (i.e. object models) This involves combining the ....

....image features and model features and choose the models with the highest counts. In all but trivial applications, however, some form of uncertainty representation (and evidence combination) is crucial to achieve robust recognition. Our approach is based on the evidential formulation described in [Shastri 1988b] Within Shastri s formulation, the problem of recognition may be viewed as decision making under uncertainty that requires choosing the most likely alternative from among a set of mutually exclusive alternatives (e.g. object models) This involves combining the evidence provided by relevant ....

[Article contains additional citation context not shown here]

L. Shastri. Semantic Networks: An Evidential Formalization and its Connectionist Realization. Research Notes in Artificial Intelligence. Morgan Kaufmann Publishers, Inc., San Mateo, California, 1988.

....It states that intelligence is best served by representing knowledge explicitly and propositionally. Connectionist researchers would, for the most part, appear not to believe in this hypothesis. In some connectionist models knowledge about the domain is represented propositionally, e.g. in Shastri s [1988] and Derthick s [1990] models, but this knowledge is compiled into the links of the network and the processing does not access the propositional representations. Most connectionist researchers appear to take the view that if the relevant aspects of the particular task are represented suitably, ....

....escape the tradeoff between computational tractability and logical soundness and completeness [Levesque and Brachman 1985] However, computational speed has always been important to connectionist researchers, and most models are built to provide fast answers, but with no guarantee of correctness. Shastri [1988] is 7 one exception; he has investigated what types of limited reasoning can be performed both soundly and correctly. Meta level reasoning Almost all connectionist models make a strong distinction between long term knowledge about how to perform the task, which is stored in the weights on the ....

[Article contains additional citation context not shown here]

Shastri, L. 1988. Semantic networks: an evidential formalization and its connectionist realization.

....Parallelism on knowledge level refers to the parallel computation of one instance of a goal concept. One possibility to exploit parallelism on this level is given by employing an isomorphic mapping between the processors of a parallel hardware and the nodes and links of a knowledge base (e.g. [9]) This turned out to be a feasible approach if both concepts and inferences are simple. In our approach, however, concepts may be complex and become a bottleneck in instantiation, since in the ERNEST formalism a concept may have an arbitrary number of attributes and relations. To get around this ....

Lokendra Shastri. Semantic Networks: An Evidential Formalization and its Connectionist Realization. Morgan Kaufmann Publishers, Pitman, London, 1988.

....is that building a full strength inference system thouroughly based on distributed representations is very difficult; on the other hand the more localized approaches stand in danger of losing the robustness that comes from distributed representations. For instance, while Shastri in his early work [14, 15] presented a system for neural knowledge representation capable of evidential reasoning on incomplete data, his newer system [1] which performs more complicated logical reasoning, only deals with crisp data. All their differences notwithstanding, a striking similarity in most of the current ....

L. Shastri, Semantic Networks: An Evidential Formalization and Its Connectionist Realization. Pitman, London, 1988.

....to the variables under consideration. As an application, we show how conceptual hierarchies, described in a high level language, can be translated into a harmony network representation, amenable to processing queries of a quite general form. In spirit this work is similar to, e.g. Shastri s [11], but whereas Shastri s networks are deterministic, and consequently require fairly complicated computing elements and control regimes, the stochastic harmony networks are structurally very simple and uniform. 2 Bayesian Networks, Harmony Networks, and Markov Random Fields We approach the problem ....

L. Shastri, Semantic Networks: An Evidential Formalization and Its Connectionist Realization. Pitman, London, 1988.

.... to the more or less heuristic models of most other hybrid systems (for our earlier, heuristic attempts towards a hybrid system, see [33, 89, 34] Secondly, although some hybrid models provide theoretical justifications for the computations (see e.g. Shastri s system for evidential reasoning [109]) they may require fairly complicated and heterogeneous computing elements and control regimes, whereas the neural network model behind our Bayesian neural system is structurally very simple and uniform, and confirms to an already existing family of neural architectures, the Boltzmann machines. ....

Shastri, L. Semantic Networks: An Evidential Formalization and Its Connectionist Realization. Pitman, London, 1988.

....the scene it is necessary to have multiple access to the shared recognizer modules. All these arguments demand for a powerful control mechanism with the representational power of a decomposition hierarchy. This in principle can also be done using ANNs [4] There are first steps in this direction [18, 16]. However, a copying mechanism for whole neural nets is extremly difficult to realize with ANNs and it can be argued that the need for a special ANN at each position of the signal is far from being convincing as an elegant mechanism, and also is very inefficient. Using semantic networks of ....

L. Shastri. Semantic Networks: An Evidential Formalization and its Connectionist Realization. Kaufmann, San Mateo, CA, 1988.

....of inferences that may be executed simultaneously. Most parallel semantic network systems employ an isomorphic mapping between the processors of a parallel hardware and the nodes and links of a knowledge base, which turned out to be a feasible approach if both concepts and inferences are simple [14]. However, since in our formalism a concept may have an arbitrary number of attributes and structural relations, complex concepts may become a bottleneck in parallel instantiation. Therfore, we employ an attribute centered representation of a semantic network, where each computation needed during ....

L. Shastri. Semantic Networks: An Evidential Formalization and its Connectionist Realization. Research Notes in Artificial Intelligence. Pitman and Morgan Kaufmann Publishers, Inc., London and San Mateo, Calif., 1988.

....reasoning using weighted links (e.g. see [22, 2] or [21] for a useful account) A key issue that remains open is the learning of appropriate control structures. We are investigating this question within the frameworks of spike timing dependent synaptic plasticity [25, 19] and recruitment learning [3, 10, 20] based on long term potentiation [6, 13] Acknowledgments This work was partially funded by NSF grants 9720398 and 9970890. ....

L. Shastri. Semantic Networks: An evidential formalization and its connectionist realization. Los Altos/Pitman Publishing Company, London, 1988.

....reasoning using weighted links (e.g. see [22, 2] or [21] for a useful account) A key issue that remains open is the learning of appropriate control structures. We are investigating this question within the frameworks of spike timing dependent synaptic plasticity [25, 19] and recruitment learning [3, 10, 20] based on long term potentiation [6, 13] Acknowledgments This work was partially funded by NSF grants 9720398 and 9970890. ....

L. Shastri. Semantic Networks: An evidential formalization and its connectionist realization. Los Altos/Pitman Publishing Company, London, 1988.

....the design of robust episodic memory modules for autonomous agents, and perhaps, to the development of memory prosthesis for brain injured humans. A few researchers have attempted the computational modeling of rapid one shot learning within a framework described variably as recruitment learning [10, 22, 8, 24, 11, 20] and vicinal algorithms [37] In simple terms, recruitment learning can be described as follows: Learning occurs within a network of randomly connected nodes. Recruited nodes are those nodes in the network that have acquired a distinct meaning (or functionality) by virtue of their strong ....

....concepts can be recruited with a high probability if one makes suitable assumptions about network connectivity. He presented a probabilistic analysis of recruitment learning based on the degree of connectivity and the number of intermediate layers in random interconnection networks. Shastri [22] extended the notion of recruitment learning to relational concepts. He treated a concept as a collection of attribute value bindings and suggested a two stage memorization process. In the first stage, binder nodes are recruited for each attribute value binding in a concept. In the second stage, ....

Shastri, L.: Semantic Networks: An evidential formalization and its connectionist realization. (see p. 181--191). Morgan Kaufmann, Los Altos/Pitman Publishing Company, London (1988).

....# p , ii) this activity is synchronous, i.e. arrives with a maximum lead lag of , iii) such synchronous activity repeats at least # times, and (iv) the interval between two successive arrivals of convergent activity is at most # iai . It has been shown [17] that recruitment learning algorithms [6, 15] proposed for one shot learning in connectionist networks can be firmly grounded in LTP. 2 A System level description of the model At a macroscopic level, the functioning of the model may be described as follows. Our cognitive apparatus construes our experiences as a stream of events and ....

L. Shastri, Semantic Networks: An evidential formalization and its connectionist realization, (Morgan Kaufamnn, Los Altos, CA, 1988).

....of faces. The primary focus of research in connectionist and neural network models has been on slow gradual learning, but some researchers have also attempted the computational modeling of rapid one shot learning within a framework described variably as recruitment learning (Feldman, 1982; Shastri, 1988; Diederich, 1989) and vicinal algorithms (Valiant, 1994) In simple terms recruitment learning can be described as follows: Learning occurs within a network of randomly connected nodes. Recruited nodes are those nodes in the network that have acquired a distinct meaning (or functionality) by ....

....conjunctive concepts can be recruited with a high probability if one makes suitable assumptions about network connectivity. He presented a probabilistic analysis of recruitment learning based on the degree of connectivity and the number of intermediate layers in random interconnection networks. Shastri (1988) extended the notion of recruitment learning to relational concepts. He treated a concept as a collection of attribute value bindings and suggested a twostage memorization process. In the first stage, binder nodes are recruited for each attribute value binding in a concept. In the second stage, ....

Shastri, L. (1988) Semantic Networks: An evidential formalization and its connectionist realization. (see p. 181--191). Morgan Kaufmann, Los Altos/Pitman Publishing Company, London, 1988.

No context found.

L. Shastri. Semantic Networks: An Evidential Formalization and Its Connectionist Realization. Pitman, London, 1988.

No context found.

L. Shastri. Semantic Networks: An Evidential Formalization and Its Connectionist Realization. Pitman, London, 1988.

No context found.

L. Shastri. Semantic Networks: An Evidential Formalization and its Connectionist Realization. Research notes in Artificial Intelligence. Pitman, London, 1988.

No context found.

L. Shastri. Semantic Networks: An Evidential Formalization and its Connectionist Realization. Pitman, London, 1988.

No context found.

L. Shastri, Semantic Networks: An Evidential Formalization and its Connectionist Realization. San Mateo, CA: Morgan Kaufmann Publishers, 1988.

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