Cohen, P., Feigenbaum, E., Eds.: The Handbook of Artificial Intelligence, vol. 3, Morgan Kaufmann, 1981.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... are reported in [19] In the relational database field various algorithms for deriving schemas in normal form have been developed to help the application builder to pin point design flaws [44] The problem of learning classes from object examples has been studied earlier in AI (e.g. 62] [16]) Clustering algorithms have been applied to build a tree of mutually exclusive classes from a given object set. Data models exist in the literature that are similar to the Demeter model used in this thesis. In particular, the notions of alternation and construction appear as classification ....

Paul R. Cohen and Edward A. Feigenbaum. The Handbook of Artificial Intelligence, volume 3. William Kaufmann, Inc., 1982.

..... Decision rule an assertion that determines the cause effect relationship between conditions and decisions. The three types of rules are useful in various applications of data mining. To enlarge the scope of their applicability, the learned rules can be generalized using the following methods [2]: replacing fixed values with intervals; e.g. rule IF a in [2, 7] THEN D = 1 is more general than the rule IF a = 3 then D = 1; deleting conjunctive elements; e.g. rule IF a = 3 THEN D = 1 is more general than the rule IF a = 3 AND b = 4 then D = 1; incorporating disjunctive elements; ....

....between conditions and decisions. The three types of rules are useful in various applications of data mining. To enlarge the scope of their applicability, the learned rules can be generalized using the following methods [2] replacing fixed values with intervals; e.g. rule IF a in [2, 7] THEN D = 1 is more general than the rule IF a = 3 then D = 1; deleting conjunctive elements; e.g. rule IF a = 3 THEN D = 1 is more general than the rule IF a = 3 AND b = 4 then D = 1; incorporating disjunctive elements; e.g. rule IF a = 3 OR b = 5 THEN D = 1 is more general than the rule ....

[Article contains additional citation context not shown here]

P. Cohen and E. A. Feigenbaum, "The handbook of artificial intelligence, " in The Handbook of Artificial Intelligence. San Mateo, CA: Morgan Kaufmann, 1983, vol. 3.

....other measures to sort the output (e.g. cross citation index, usage frequency, user recommendations . 2. Show additional information to enable the user to browse the list faster. In computer vision labelling techniques and algorithms have a long tradition for constraint propagation (e.g. Cohen, Feigenbaum (Eds. 1982)) or for labelling image parts (e.g. Leung, Yang (1995) However, in this paper we understand labelling as a visualization technique for Internet resources. Section 2 describes the labelling process. Next, we discuss several dimensions of labelling like time, region, language, rating and resource ....

COHEN, P.R. and FEIGENBAUM, E.A. (Eds.) (1982): The Handbook of Artificial Intelligence, Vol. 3, Chapter XIII: Vision, William Kaufmann, Los Altos, California.

....salientcontour measure wewant should tell us the likelihood of an image contour at a given image position and orientation, based on the responses of oriented filters. It should favor long, straight curves, and continue over gaps. Anumber of approaches could be used, including relaxation labelling [24, 52, 44, 83], the approach of Grossberg and Mingolla [39, 40] snakes [53] or dynamic programming [95] Splines [11] or elastica [81, 80] could be used to interpolate across gaps. We implemented the dynamic programming method of Shaashua and Ullman [95] It solves an explicit 87 optimization problem, gives ....

P.R.CohenandE.A.Feigenbaum, editors. The Handbook of Artificial Intelligence, pages 292--300. Addison Wesley,1982.

....of using junction information to interpret scenes. 5.1 Related Work 5.1.1 Blocks World Vision researchers studying the blocks world developed important methods for using local information to interpret scene structure. The blocks world restricts scene objects to be polyhedral blocks. See [23] for a review) Guzman [42] made use of vertices and junctions to recognize 3 dimensional objects. He developed heuristics for grouping the elements of a line drawing into objects. Huffman [51] and Clowes [22] systematically labelled each line as corresponding to either a concave edge, a convex ....

P.R.CohenandE.A.Feigenbaum, editors. The Handbook of Artificial Intelligence, pages 139--194. Addison Wesley,1982. 125

....language processing, symbolic programming languages, expert systems, planning, vision, and deduction systems were all widely pursued but with a distinctly symbolic flavor and to the exclusion of connectionist techniques. This is punctuated by the fact that Cohen and Feigenbaum s classic textbook [10], largely considered the bible of the field, devotes a mere five pages out of three volumes to perceptron like learning techniques. Rumelhart and McClelland [53] are largely responsible for the renewed popularity of multilayer perceptrons that existed in the mid 1980 s. While they were not the ....

P.R. Cohen and E.A. Feigenbaum, editors. The Handbook of Artificial Intelligence, volume 3. AddisonWesley Publishing Company, Inc., 1982.

....The text is often several factors shorter than a complete object composition description, yet it contains the same information. Class dictionary graphs are an abstraction of semantic networks with part of and is a links. Semantic networks have been studied extensively in the AI literature [7]. Class dictionary graphs are also related to AND OR graphs which have been used in AI [7] The Interface Description Language described in [33] uses structures related to class dictionary graphs. However, the specific axiomatic structure which we introduce here as well as our efficient axiom ....

....yet it contains the same information. Class dictionary graphs are an abstraction of semantic networks with part of and is a links. Semantic networks have been studied extensively in the AI literature [7] Class dictionary graphs are also related to AND OR graphs which have been used in AI [7]. The Interface Description Language described in [33] uses structures related to class dictionary graphs. However, the specific axiomatic structure which we introduce here as well as our efficient axiom checking algorithm are new, to the best of our knowledge. The relationships between this ....

Paul R. Cohen and Edward A. Feigenbaum. The Handbook of Artificial Intelligence, volume 3. William Kaufmann, Inc., 1982.

....literal descriptions of simple movements, compared with the complex descriptions and interpretations produced by humans. 2.3. 1 Levels of Processing in Computer Vision Systems Computer vision has been defined as the information processing task of understanding a scene from its projected images (Cohen and Feigenbaum 1982:127) Computer vision systems take low level codings of still and moving images as input and produce higher level, symbolic classifications and descriptions of the objects and actions depicted by the pixels. Thus, research in computer vision has been characterised as following a signals to ....

Cohen and Feigenbaum (1982). Paul R. Cohen and Edward A. Feigenbaum, The Handbook of Artificial Intelligence. Reading MA: Addison-Wesley.

.... evolution, it was still required to endow the teacher role with some knowledge of the control actions needed (e.g. see [Widrow, et al., 1973] Continuing up into the early 1980 s, AI researchers who utilized critics were also typically obliged to provide the critic with domain specific knowledge [Cohen Feigenbaum, 1982]. In 1983, however, a very important extension was made in which the critic also learns, in this case, to provide increasingly better evaluation feedback; this removed the requirement for providing the teacher a priori knowledge about desired controller actions [Barto, et al., 1983] Historically, ....

Cohen, P.R. & E.A. Feigenbaum, 1982, The Handbook of Artificial Intelligence, vol 3, Kauffman.

.... there are cases in which on line learning is strictly harder than pac learning [B] 3 The Credit Assignment Problems The credit assignment problem may be defined as the problem of assigning credit or blame to the individual decisions that led to some overall results (Cohen and Feigenbaum [CF]) Obviously, this problem is ubiquitous not only in Artificial Intelligence, but also in the study of adaptive neural networks, where credit or blame for the overall performance of the network has to be distributed to the individual components of the network. In the study of learning unions of ....

P. Cohen, E. Feigenbaum, "The Handbook of Artificial Intelligence", Vol. 3, William Kaufman, 1982. 34

....a machine learning approach, using manually structured example portions to generate recognition grammars for automatically structuring the rest of the document. In this paper we present MarkItUp , a system which uses techniques for editing by example [12,13] and more generally learning by example [14,15] to gradually acquire recognition grammars. MarkItUp supports a structure editor which can be used to transform example portions of documents into SGML documents by inserting markups. Predefined recognition patterns, such as delimiters or application specific formats, are used to abstract from ....

P.R. Cohen and E.A. Feigenbaum, The handbook of artificial intelligence, volume 3, William Kaufmann, Los Altos, CA, 1982.

....and across displays, and unity addresses how to unify individual components into a coherent whole [14] In our approach, the design of a visual discourse is a sequence of visual actions. A visual action is an encoded graphic design technique (e.g. move an object) We use a hierarchical planner [4] to generate the sequence of actions. Compared to search based approaches, hierarchical planning achieves computational efficiency by reducing the amount of search needed [24] It also eases the task of knowledge encoding [20] by reusing the actions that are common to many design tasks. In ....

P. Cohen and E. Feigenbaum, editors. The Handbook of Artificial Intelligence, volume 3, chapter XV. Planning and Problem Solving. Addison -Wesley, Reading, MA, 1982.

....in mind, a typical fit and splitassignment problem can be characterized by describing a problem generator. The generator is listed in figure 4. 2. 4 Related Learning Problems The fit and split assignment problem is a specialization of the overlapping concept learning problem [Michalski, 1983; Cohen and Feigenbaum, 1982] where a class corresponds to a partial hypothesis. It differs from standard overlapping concept learning in two ways. First, examples are labeled with every class (that is, partial hypothesis) that covers them. In the standard problem, examples are labeled with only one covering class at a ....

P. Cohen and E. Feigenbaum. The Handbook of Artificial Intelligence. Volume 3, HeurisTech Press, Stanford, CA, 1982.

....the polymorphic method as an abstract concept. We are actively pursuing the formalization of these ideas at the present time. 4 Related work The fast algorithm which we present here has not been reported in the literature. Our work is a continuation of earlier work on inductive inference [CF82, Chapter XIV: Learning and inductive inference] AS83] Our contribution is an efficient algorithm for inductive inference 42 of high level class descriptions from examples. Related work has been done in the area of learning context free grammars from examples and syntax trees [AS83] The key ....

.... reported in [DJ88] In the relational database field various algorithms for deriving schemas in normal form have been developed to help the application builder to pin point design flaws [Lie85] The problem of learning classes from object examples has been studied earlier in AI (e.g. SM86] CF82] Clustering algorithms have been applied to build a tree of mutually exclusive classes from a given object set. Our work extends this earlier work since we have more structure in our classes, e.g. the capability to define a language. The axiomatic Demeter kernel model which is used in this ....

Paul R. Cohen and Edward A. Feigenbaum. The Handbook of Artificial Intelligence, volume 3. William Kaufmann, Inc., 1982.

....choices for action. At any point in time, they can choose to act on the basis of the constraints they have recalled and examined thus far, or they can extend 78 Thanks to Vincent Wright for introducing me to the term waffling. 79 e.g. PLANNER, CONNIVER, HACKER, etc. Barr and Feigenbaum, 1981; Cohen and Feigenbaum, 1982]. Chapter 5: A Constraint Directed Approach to Improvisation 169deliberation, allowing more information to accumulate (and potentially, opportunities to act to pass by) Choosing to act includes not only commitment to physical actions, but also to cognitive actions, such as adopting or ....

Cohen, Paul R., and Edward A. Feigenbaum (Eds.), The Handbook of Artificial Intelligence, Volume III (Los Altos, CA: William Kaufmann), 1982. 639 pp.

....the compressed fuel and air mixture is ignited and burns. Consequently, a teleological description will associate a modification with a specification of the nearest structural parent (hierarchically) within which the modification is made. 9. 4 Planning Planning systems (cf. Cohen and Feigenbaum [CF82], Nilsson [Nil80] including linear planning, nonlinear planning (cf. Fikes, Nilsson [FN71] and Hart [FHN72] and hierarchical planning (cf. Sacerdoti [Sac74, Sac77] provide additional examples of initial and evolved specifications in design. Planning systems attempt to achieve some goal state, ....

Paul R. Cohen, Edward A. Feigenbaum, The Handbook of Artificial Intelligence, Vol. III, Addison-Wesley, Reading, MA.

....of logic. AI researchers sought to encode knowledge about problem solving in the form of explicit, or declarative, representations of objects and relationships in the world, in constrast to the traditional procedural approach as typified by OR solution methodologies (Barr and Feigenbaum 1982; Cohen and Feigenbaum 1982) AI researchers also dismissed numerical methods as relatively unimportant to decisions made by agents with the abilty to manipulate abstract symbols. In particular, methods for optimizing the expected value of action under uncertainty seemed inadequate for explaining cognition and intelligent ....

Cohen, P. and Feigenbaum, E., The Handbook of Artificial Intelligence, volume 2. William Kaufmann.

....programmer. in other words, A programmer has executed an editor program. B. Validity of generalized tuples From the viewpoint of inductive learning, a given database(or a part of it) and a set of possible generalized tuples are regarded as an instance space and a pattern space, respectively[8]. Our summarization process searches a pattern space to choose valid generalized tuples with respect to a given instance space, i.e. a given database. Recall that the goal of database summarization is to generate representative descriptions embracing as many database tuples as possible. Thus, the ....

P.Cohen and E.Feigenbaum, The Handbook of Artificial Intelligence, Vol.3, William Kaufmann Pub., 1982, pp. 411-415

....editing. The inference mechanism determines constants in the constraint equations, but it does not synthesize new classes of equations. Our technique is an application of learning from multiple examples, also known as empirical learning. Several empirical learning systems are discussed in [Cohen82] In contrast, generalizing from a single example is called explanation based learning and is surveyed in [Ellman89] Explanation based learning requires a potentially large amount of domain knowledge to determine why one explanation is particularly likely. As we illustrate in subsequent ....

Cohen, Paul R., and Feigenbaum, Edward A. The Handbook of Artificial Intelligence. vol. 3. Kaufmann, Inc., Los Altos, CA. 1982.

....in section 2. The main ingredient of this learning algorithm is a novel solution of the credit assignment problem . The credit assignment problem may be defined as the problem of assigning credit or blame to the individual decisions that led to some overall result (Cohen and Feigenbaum [CF]) Obviously this problem is ubiquitous not just in Artificial Intelligence, but also in the study of adaptive neural networks, where credit or blame for the overall performance of the network has to be distributed to the individual components of the network. The credit assignment problem in the ....

P. R. Cohen, E. A. Feigenbaum, "The Handbook of Artificial Intelligence" vol. 3, William Kaufmann (Los Altos, 1982).

....his or her reality. In this process, the logician is guided by principles that are mainly theory dependent. Formal principles, such as completeness or soundness; aesthetic 1 See (Newell and Simon, 1956) or (Simon, 1991) for very early logical motivations and concerns about automated learning; (Cohen, et al. 1982), Chpt. XIV, for a classical AI overview; Michalski et al. 1983) for seminal papers, and (Carbonell, 1990) and (Kodatroff and Michalski, 1990) for recent works and overviews. principles of elegance, economy and simplicity; and, pragmatic principles like the quality and scope of the ....

.... are condensed into a (generally) much reduced set of sentences from which all the examples can be deduced (for example, the general diagnostic rules 3 Other, more elaborated, definitions of machine learning or automated learning have been proposed by the AI community, see for example (Cohen et al. 1982) pp. 324 ss. Carbonell, 1990) pp 1 9; Anderson, 1990) or (Michalski and Kodatroff, 1990) are induced from a database containing simple clinical reports) Quinlan, 1983; Michalski and Chilausky, 1980; Goldberg, 1989) Second, there are works that relax the truth conditions of the theories, ....

Cohen, P. and Feigenbaum, E. (1982): Handbook of Artificial Intelligence, Vol. 3. HeurisTech Press & Morgan Kaufmann, Palo Alto, CA.

....of the system at that time. A problem can be represented in a number of ways using that basic convention. Neuromorphic Distributed General Problem Solvers, Andrzej Bieszczad 18 Systems and Computer Engineering, Carleton University 3.1. 1 State spaces In the state space approach (Feigenbaum [22], Nillson [41] Winston [64] Russel [48] each state is a point in an n dimensional space with the n coordinates corresponding to every observable parameter of the system. At each point in time, the system is in the state that is being represented by a point, or a vector drawn from the origin to ....

....idea what the other representations might be. j i ( d c i S m , d c j S m , Systems and Computer Engineering, Carleton University Neuromorphic Distributed General Problem Solvers, Andrzej Bieszczad 21 3.1.2. 1 Problem reduction In the problem reduction approach (Feigenbaum [22]) a problem is represented by a triple S, O, P , where: S a set of initial states, O a set of reduction operators and . P a set of primitive problems that can be resolved. Using the problem reduction representation, the problem is solved when it is reduced to a number of primitive ....

[Article contains additional citation context not shown here]

Feigenbaum, E. A. et al. (1987), Handbook of Artificial Intelligence, Morgan Kaufman, Los Altos, CA.

....corresponds to an algorithm schema plus a list of search control rules and mechanism flags. The space of possible configurations is exponential in the number of available control rules and flags. Since this space is much too large to search exhaustively, Multitac employs a beam search (Cohen and Feigenbaum, 1982), a form of parallel hill climbing that heuristically searches only a small portion of the space. 13 The beam search takes a beam width B, a set of training instances, and an instance time bound T . The goal is to find the configuration that performs best on the training instances. The best ....

P. R. Cohen and E. A. Feigenbaum, editors. The Handbook of Artificial Intelligence, Volume III, volume Volume 3. William Kaufmann, Inc., Los Altos, California, 1982.

.... path from the start node to the goal node in the expert diagram using the cost function: f(n) g(n) h(n) where g(n) is the distance of the path from the current node n in the graph back to the start node, and h(n) is a heuristic estimate of the distance from the current node n to the goal [3]. The heuristic is articulated as follows: If the student has indicated a FOR link, all paths in the expert diagram which contain AGAINST links will be given shorter distances than paths with FOR links. Likewise, if a student has indicated an AGAINST link, all paths in the expert diagram which ....

P R Cohen and E A Feigenbaum. The Handbook of Artificial Intelligence, volume 1. AddisonWesley, New York, 1989.

....statements (see Gordon and Subramanian, 1994) In many task domains, the advice giver may wish to use natural, but imprecise, terms such as near and many. A compiler for such terms will be needed for each general environment since the terms needed to describe an 1 See also pg. 345 349 of Cohen and Feigenbaum (1982). problem will be problem dependent. In ratle, we make use of a method similar to Berenji and Khedkar s (1992) for compiling fuzzy logic terms into neural networks. Step 4. Integrate the reformulated advice into the agent s current knowledge base. In this work our agent employs a connectionist ....

Cohen, P. & Feigenbaum, E. (1982). The Handbook of Artificial Intelligence (volume 3).

....Our method yields results identical to those of methods based on interleaving semantics, it just avoids most of the associated combinatorial explosion. This method can be qualified as nonlinear though it differs substantially from classical nonlinear backward search methods such as NOAH [Cohen and Feigenbaum, 1982], SIPE [Wilkins, 1984] TWEAK [Chapman, 1987] Our search method is based on recent results in concurrent system verification. In [Godefroid, 1990] it is shown that the global behavior of a set of communicating processes can be represented by an automaton which can be much smaller than the usual ....

.... and yields a successor state S 0 such that S 0 = S Gamma pre(a) post(a) what disables all the preconditions of the action and enables all its effects (note that the set of preconditions and the set of effects of an action may be nondisjoint) There are other ways to represent actions [Cohen and Feigenbaum, 1982]. Just note that every planning problem expressed in these frameworks can be encoded in the framework presented here. Anyway, the results of this paper could be adapted to other action representations. For convenience, we assume that the goal is a conjunction of facts describing totally the state ....

P. R. Cohen and E. A. Feigenbaum. Handbook of Artificial Intelligence. Pitman, London, 1982.

....objects. The text is often several factors shorter than a complete object composition description, yet it contains the same information. Class dictionary graphs are an abstraction of semantic networks with part of and is a links. Semantic networks have been studied extensively in the AI literature [9]. Class dictionary graphs are also related to AND OR graphs which have been used in AI [9] The Interface Description Language described in [35] uses structures related to class dictionary graphs. However, the specific axiomatic structure which we introduce here as well as our efficient axiom ....

....yet it contains the same information. Class dictionary graphs are an abstraction of semantic networks with part of and is a links. Semantic networks have been studied extensively in the AI literature [9] Class dictionary graphs are also related to AND OR graphs which have been used in AI [9]. The Interface Description Language described in [35] uses structures related to class dictionary graphs. However, the specific axiomatic structure which we introduce here as well as our efficient axiom checking algorithm are new, to the best of our knowledge. The Format model by Hull and Yap ....

Paul R. Cohen and Edward A. Feigenbaum. The Handbook of Artificial Intelligence, volume 3. William Kaufmann, Inc., 1982.

....is intended to provide readers unfamiliar with machine learning a broad view of the different approaches. Readers who want to learn more about these techniques are directed to the references cited in respective sections. Introductory readings in machine learning can be found in (Carbonell, 1989; Cohen and Feigenbaum, 1982; Michalski et al. 1983) 3.1 Inductive Learning Inductive learning (also known as empirical learning or similarity based learning) involves forming concepts by finding similarities and differences among data. This approach is best suited for tasks where a considerable amount of data is ....

Cohen, P. and Feigenbaum, E., editors (1982). The Handbook of Artificial Intelligence, volume 3, chapter XIV, pages 323--511. William Kaufmann, Los Altos, CA.

....its prediction. In connectionist Qlearning, the utility function is implemented as a neural network, whose inputs describe the current state and whose outputs are the utility of each action. We now return to the task of advice taking. HayesRoth, Klahr, and Mostow (1981) also see pg. 345 349 of Cohen Feigenbaum 1982) described the steps involved in taking advice. In the following subsections, we state their steps and discuss how we propose each should be achieved in the context of RL. Step 1. Request the advice. Instead of having the learner request advice, we allow the external observer to provide advice ....

Cohen, P., & Feigenbaum, E. 1982. The Handbook of Artificial Intelligence, Vol. 3. William Kaufmann.

....3.1.4.1 Postprocessing Using Statistical Language Model Word collocation is a simple but powerful statistical language model. By formalizing the FRAMEWORK 42 word candidate selection problem as an instance of a constraint satisfaction problem [97] we proposed relaxation(pp. 292 300 in [29]) for word candidate selection. The probabilistic relaxation algorithm uses word collocation trained from large text corpora to re evaluate the confidence scores of word candidates and to re rank them based on their new confidence scores. The basic idea of the relaxation algorithm is the use of ....

P.R. Cohen and E.A. Feigenbaum, editors. The Handbook of Artificial Intelligence, Volume 3. William Kaufmann, INC., 1982.

....one for each disease category. It considers all training instances of a particular category to be positive examples and all other training instances to be negative ones. The AQ11 program is nearly equivalent to repeated application of the candidate elimination approach to concept learning [Cohen Feigenbaum, 1982], and to convey the general idea of how it works we will briefly sketch that method. The candidate elimination algorithm presupposes a language in which objects are expressed (e.g. a 49 element vector of integers in the soybean case) a language in which descriptions are expressed (e.g. a ....

Cohen, P.R. and Feigenbaum, E.A. (Editors), 1982: The Handbook of Artificial Intelligence, Vol III. William Kaufmann, Los Altos, CA, pp. 423--427.

....editing. The inference mechanism determines constants in the constraint equations, but it does not synthesize new classes of equations. Our technique is an application of learning from multiple examples, also known as empirical learning. Several empirical learning systems are discussed in [Cohen82]. In contrast, generalizing from a single example is called explanation based learning and is surveyed in [Ellman89] Explanation based learning requires a potentially large amount of domain knowledge to determine why one explanation is particularly likely. As we illustrate in subsequent ....

Cohen, Paul R., and Feigenbaum, Edward A. The Handbook of Artificial Intelligence. vol. 3. Kaufmann, Inc., Los Altos, CA. 1982.

....graph structure in Figure 1; the absence of an edge means that the vertices are not related by that relation (e.g. the two hydrogens are not bonded) For brevity, negated angular relationships are omitted from the figure. many machine learning systems rely on the single representation trick (Cohen and Feigenbaum, 1982) by representing objects as specially identified individual concepts. In many terminological knowledge representation systems (Nebel, 1990; MacGregor and Brill, 1992) instantiation is also computed as subsumption between a concept and an individual concept. The subsumption problem is known to ....

....a concept abstraction or generalization operator (Chang and Liu, 1984) The notion of a structured concept reappears frequently in many different problem domains. The structural chemistry community, for example, has had a longstanding interest in graph structures. An early system, Meta DENDRAL (Cohen and Feigenbaum, 1982), created a taxonomy of molecular graphs for use in mass spectrum analysis. The system of Okada and Wipke (1989) indexes a database of molecular graphs using a taxonomy of discovered maximal common subgraphs. Okada (1993) shows how this taxonomy can be used for analogical inference. Structurally ....

Cohen, P. R. and Feigenbaum, E. A. 1982. The Handbook of Artificial Intelligence, volume 3. William Kaufmann.

....typical ICAI system. Having been through this process, the author will present in this paper this latter perspective so that the designers of future ICAI systems can readily leverage the lessons learned by their predecessors. The interested reader is referred to Fletcher [19] Feigenbaum and Barr [18], and Park et al. 40] for good surveys of past ICAI systems, and to Park et al. 40] for an extensive comparison of CAI and ICAI. II Learning Scenarios A learning scenario is the situation in which the student s learning is to take place. The most important criterion with which to view a ....

....intersections between two concepts in the data base. scholar uses a similar intersection search to answer questions like Is Buenos Aires in Brazil scholar can determine what two things have in common by searching both their superp (superpart) or superc (superconcept) links for an intersection [18]. This allows scholar to relate a student s answer to the correct one and thus realize that if a city with certain characteristics was asked for, the student is not as far off if his answer is an incorrect city as if he answered with some country [13] There are many other techniques for ....

[Article contains additional citation context not shown here]

E.A. Feigenbaum and A. Barr, Eds., The Handbook of Artificial Intelligence, Vol. 2. Los Altos, CA: William Kaufmann Inc., 1982, Chapter 9.

....a machine learning approach, using manually structured example portions to generate recognition grammars for automatically structuring the rest of the document. In this paper we present MarkItUp , a system which uses techniques for editing by example [12,13] and more generally learning by example [14,15] to gradually acquire recognition grammars. MarkItUp supports a structure editor which can be used to transform example portions of documents into SGML documents by inserting markups. Predefined recognition patterns, such as delimiters or application specific formats, are used to abstract from ....

P.R. Cohen and E.A. Feigenbaum, The handbook of artificial intelligence, volume 3, William Kaufmann, Los Altos, CA, 1982.

....circle or small square . In this case, there are multiple solution paths through the learning tree. In this case, it would be beneficial to be able to find the optimal path to any solution; this is a major goal of ID3. 3.1. 2 CLS Implementation This section outlines the CLS algorithm as given in [3]. Like most algorithms in any science, CLS has differing implementations depending on the final user s requirements; this document presents a general case suited to the examples presented. Define I to be a set of training instances. This set can be a random sampling of the full data set, or it ....

....to Figure 3, the original set of colored shapes is I ; after one iteration, the subsets are divided into two windows, based on their similarities to the feature F, shape . Each window is then divided into smaller windows based on the second level F, size . The ID3 algorithm abstract as given in [3] follows. 1. Randomly select a window from I . 1 2. Use CLS to form a rule regarding the structure of the decision tree produced by the window. 3. Iterate through the entire set I to find exceptions to the rule produced. 4. Form a new window by combining the current window with the ....

Cohen, Paul R, et. al. The Handbook of Artificial Intelligence, Volume 3. pages 406--411. HeurisTech Press, Stanford, CA, 1982.

No context found.

Cohen, P., Feigenbaum, E., Eds.: The Handbook of Artificial Intelligence, vol. 3, Morgan Kaufmann, 1981.

No context found.

Cohen, P. R. and Feigenbaum, E. A. 1982. The Handbook of Artificial Intelligence, volume 3. Kaufmann.

No context found.

P.R. Cohen and E.A. Feigenbaum, editors. The Handbook of Artificial Intelligence, volume 3. Morgan Kaufmann, 1981.

No context found.

Bar A. and Feigenbaum E. A. (1981) The Handbook of Artificial Intelligence, Kaufman.

No context found.

Paul R. Cohen and Edward A. Feigenbaum. The Handbook of Artificial Intelligence, volume 3. William Kaufmann, Inc., 1982.

No context found.

Cohen, P. and Feigenbaum, E.A., The Handbook of Artificial Intelligence 3 (Kaufmann, Los Altos, CA, 1982).

No context found.

Paul R. Cohen Avon Barr and Edward A. Feigenbaum. The Handbook of Artificial Intelligence, volume 4. Addison Wesley, Reading, MA, 1989.

No context found.

Baar, A., Cohen, P.R,. Feigenbaum, E.A.: The handbook of artificial intelligence 4 Addison Wesley (1989)

No context found.

P. R. Cohen and E. A. Feigenbaum. Handbook of Artificial Intelligence, Chap. XIII. William Kaufman, Los Altos, CA, 1982.

No context found.

Paul R. Cohen and Edward A. Feigenbaum, eds., Handbook of Artificial Intelligence, vol. 3, William Kaufmann Inc., 1982, pp. 313-321. 125

No context found.

Paul R. Cohen and Edward A. Feigenbaum. The Handbook of Artificial Intelligence, volume 3. HeurisTech Press, 1982.

No context found.

Paul R. Cohen and Edward A. Feigenbaum. The Handbook of Artificial Intelligence, Vol. III. Addison-Wesley Pub. Co., Reading, Massachusetts, 1982.

No context found.

P.R. Cohen and E.A. Feigenbaum, editors. The Handbook of Artificial Intelligence, volume III. Addison-Wesley Publishing Co., Reading, MA, 1986.

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