P.H. Winston. Learning new principles from precedents and exercises. Artificial Intelligence, 19(3):321--350, 1982.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....details can be found in [35] For this paper, the key thing to note is that the analogy ontology successfully enabled KRAKEN to use analogical processing facilities, as intended. Related Work There have been a number of systems that capture some aspects of reasoning by analogy. Winston [39] describes a system that extracts rules from precedents, but was only tested with very small (10 or so propositions) examples. Special purpose matchers and retrievers have been used for exploiting analogy in problem solving and planning (cf. 1,37] but such systems lack the flexibility to deal ....

Winston, P. 1982. Learning New Principles from Precedents and Exercises," Artificial Intelligence 19, 321350

....for solving geometric analogy intelligence test questions. In 1975, Simon pointed out that design and creation is a class of problems featured by their synthesis nature [6] In early 1980s, Winston published his pioneering results on the relationship between learning, reasoning and analogy [24, 25]. Other fundamental work on learning by analogy includes [12] and [19] Holyoak concluded that analogical thinking is an important feature of human intelligence [16] Keane [20, 21] applied analogical mechanisms to problem solving. Our approach is also based on analogical reasoning. We devised a ....

P. H. Winston, Learning new principles from precedents and exercises, Artificial Intelligence, 19:321, 1982.

....incrementally. Particularly relevant to our approach is Winston s work on analogy [52] which focused attention on the di erences new knowledge has from what is already known, and, in consequent work, processed English like input to be able to draw analogous inferences given similar new input [53]. Also relevant notion of analogy in Teiresias, and Forbus s approach to analogy by mapping between partially aligned structures of concepts in two domains, as formalized in the work on the Structure Mapping Engine (SME) 17] 2.3 Knowledge representation We brie y overview several systems that ....

....Sections 4.6 and 4.7 elaborate on the generators and critics we propose to implement. 4.6 Generators: analogy and inference We use two classes of mechanisms for generating questions. The rst is analogy based and the second one is inference based. We propose analogical learning (see also [23, 53]) as a key method for handling noisy, poorly integrated input. The key new feature of the analogy algorithm we propose is to map from many similar objects, obtaining a much cleaner sum analogy . Because of its reliance on many similar object, we call the method k nearest neighbor analogy or ....

P. H. Winston. Learning new principles from precedents and exercises. Arti cial Intelligence, 19(3):321-350, 1982.

....B follows, and that A 0 correspond to A, analogically conclude B 0 . As rst stated by Peirce [14] research on analogy can be divided into two directions. The rst one (that we call replacement based analogy) de nes analogy as a replacement of the source object with the target object [19], 10] 11] 8] The other one (that we call determination based analogy) de nes analogy as a generalization from one instance [6] 2] 9] Approaches to analogical reasoning in AI have been concerned mainly with the cog S. Costantini and G.A. Lanzarone Analogy in Logic Programming 2 ....

....terms of possible further extensions to the approach, which naturally arise in the context of DIANA. In Section 7 we conclude with comparisons to related work and open problems. 2 Informal Description of the Approach Replacement based analogy is based on the following principle, due to Winston [19] and reported in [8] Assume that the premises 1 ; n logically imply in the source domain. Assume also that the analogous premises 0 1 ; 0 n hold in the target domain. Then we conclude atom 0 in the target which is analogous to . In [8] the source and target ....

P. Winston, Learning New Principles from Precedents and Exercises, Articial Intelligence, 19 (1982), 321-350.

....In contrast, in our theory, abstract models are useful in both the access and transfer stages of analogical reasoning. Moreover, in our approach learning is not only failure driven but it also occurs from successful experiences. Learning Method: Our model based approach to learning is similar to Winston s model (1982) which shows that learning can be done by analogically transferring causal links in the explanation of an example to the target concept. Our approach is also similar to explanation based methods such as EBG (Mitchell, Keller, KedarCabelli, 1986) and EBL (DeJong Mooney, 1986) in using ....

Winston, P.H. 1982. Learning New Principles from Precedents and Exercises. Artificial Intelligence 19(3):321-350.

....analogy. Given a plan that succeeds for the base, we apply the plan to the target and propose to correct its failures by an abductive recovery mechanism inspired from abductive recovery from failed proofs. 1 Introduction The analogy scheme we shall use in this paper is quite a classical one (Winston, 1982; Gentner, 1983; Chouraqui, 1985; Falkenhainer, Forbus, and Gentner, 1986; Carbonell, 1983, 1986; Kedar Cabelli, 1988; Kodratoff, 1988) It can be described as follows. Let us suppose that we dispose of a piece of information, the base, that can be put into the form of a doublet (A, B) in which it ....

Winston, P. H. Learning new Principles from Precedents and Exercises, Artificial Intelligence 19, 1982, pp. 321-350.

....explanations of specific experiences dates at least as far back as Winston [1980] Winston s model assumed knowledge of what is the concept being learned and relied on information concerning whether an example is a positive instance or negative instance of the concept. Our approach is similar to Winston s later models [1982; 1986] that show that learning can be done by analogically transferring causal links in the explanation of an example to the target concept. Our approach is also similar to explanation based learning (EBL) DeJong and Mooney, 1986; Mitchell et al. 1986] in using explanations (SBF models) to ....

P. Winston. Learning New Principles from Precedents and Exercises. Artificial Intelligence, 19(3):321--350, 1982.

....incomp(x i jB) incomp(x i ) Gamma u. A signi cant increase of performance must take place between the rst sequence and the later sequences. The parameters are the same as the rst case, the complexity of B and the constant u. This learning from precedents has also been studied in AI (see e.g. Winston 1982). Knowledge Applicability may also be correlated with deductive abilities and these may also correlate with the idea of congruence or coherence, since TT JHdez.tex; 20 05 1999; 11:10; p.16 Beyond the Turing Test 17 it can be measured as constraint satisfaction (Thagard 1989) Other factors are ....

P. H. Winston. Learning New Principles from Precedents and Exercises. Artiøcial Intelligence, 19(3), 1982.

....in P 2 , and fi i and fi 0 i are analogous for each i (1 i n) Then analogical reasoning is to derive a ground atom ff 0 in P 2 which is analogous to ff. We follow the principle of analogical reasoning by Haraguchi and Arikawa [7] shown in Figure 2, that is based on Polya [14] and Winston [18, 19]. P 1 : fi 1 ; fi n ff m m (similarity ) P 2 : fi 0 1 ; fi 0 n ff 0 Figure 2: The principle of analogical reasoning. The in Figure 2 denotes an analogy which gives a similarity between P 1 and P 2 . In this paper, we take an analogy as a partial identity as defined ....

P. H. Winston. Learning new principles from precedents and exercises. Artificial Intelligence, Vol. 19, pp. 321--350, 1983.

....approach described here in targeting the combined causal reconstruction causal modeling problem occurring when a program is simultaneously presented with a causal description and a demonstration of a target physical behavior. The work described here is also related to research in analogy (e.g. [10, 24, 64]) and possible interactions arise in this context as well. As many analogical reasoning programs represent individual events by the equivalent of atomic formulae (e.g. COLLIDE(o1, o2) these fitting into larger systems of events to be associated analogically, a promising approach would ....

P. H. Winston, Learning new principles from precedents and exercises, Artif. Intell. 19 (3) (1982) 321--350.

....well known knowledge in the base domain on the target domain by using the analogy. Thus, it is an essential point for analogical reasoning to compute an analogy which maps from a base domain to a target domain. Then, Many authors have extensively studied analogical reasoning from this point of vew [1, 5, 6, 11, 16, 17]. However, there often arises a problem of combinatorial explosion in computing analogies [3] 3 This work is partly supported by Grant in Aid for Scientific Research on Priority Areas from the Ministry of Education, Science and Culture, Japan. y JSPS Fellowship for Japanese Junior ....

....1 ; fi n hold in P 2 , and ff i and fi i (1 i n) are analogous. Then, analogical reasoning is to derive a fact fi in P 2 which is analogous to ff. Thus, we follow the principle of analogical reasoning by Haraguchi and Arikawa [6] shown in Figure 1 which is based on Polya [11] and Winston [16, 17]. P 1 : ff 1 ; ff n ff m similarity m P 2 : fi 1 ; fi n fi Figure 1: The principle of analogical reasoning. Then, in Figure 1 denotes an analogy which gives a similarity between the base and the target. Analogical reasoning is carried out by projecting some of the base ....

P. H. Winston. Learning new principles from precedents and exercises. Artificial Intelligence, Vol. 19, pp. 321--350, 1983.

....Forbus, and Gentner 1990) based upon the structure mapping theory of Dedre Gentner (1983) We will examine this model within the context of our earlier remarks. Other models of analogy making, such as those of Burstein (1986) Carbonell (1986) Holyoak Thagard (1989) Kedar Cabelli (1988) and Winston (1982), while differing in many respects from the above work, all share the property that the problem of representation building is bypassed. Let us consider one of the standard examples from this research, in which the SME program is said to discover an analogy between an atom and the solar system. ....

....The work of Kedar Cabelli (1988) takes a limited step in this direction by employing a notion of purpose to direct the selection of relevant information, but still starts with all representations pre built. Other researchers, such as Burstein (1986) Carbonell (1986) and Winston (1982), all have models that differ in significant respects from the work outlined above, but none of these addresses the question of perception. The ACME program of Holyoak and Thagard (1989) uses a kind of connectionist network to satisfy a set of soft constraints in the mapping process, thus ....

Winston, P. H. (1982). Learning new principles from precedents and exercises. Artificial Intelligence 19: 321-350.

....Psychological data indicates that people often use analogies in solving problems of various kinds [Gick and Holyoak, 1980, 1983; Gentner, 1983] Computational models of analogy date as early as Kling [1971] though these early models were only approximate. Later computational models such as [Winston, 1982; 1986] Gentner, 1983; Falkenhainer et al. 1989] Kedar Cabelli, 1988] Carbonell, 1986] and [Holyoak and Thagard, 1989] are more complete but still address only some stages of AR, mostly the mapping (or transfer) stage. For instance, SME [Gentner, 1983; Falkenhainer et al. 1989] PDA ....

.... the source domain and not their attributes [Gentner, 1983] Most theories of analogical reasoning also involve transferring relationships directly from a source analog to the target problem: relationships that conform to systematicity principle as in [Gentner, 1983] or causal relationships as in [Winston, 1982; 1986] or functional relationships as in [KedarCabelli, 1988] are transferred. Our work is similar to both [Winston, 1982; 1986] and [KedarCabelli, 1988] in that we also transfer causal and functional relationships in a source domain to the target domain, although the tasks we model are quite ....

[Article contains additional citation context not shown here]

P. Winston. Learning New Principles from Precedents and Exercises. Artificial Intelligence, 19(3):321--350, 1982.

....from the literature, and its output was congruent with empirical findings. 1 Introduction 1.1 The Stages of Analogy Many new theories of complex analogy have been advanced this decade. Some of the best known are Gentner s (1983) structure mapping theory, Holyoak s (1985) pragmatic theory and Winston s (1980, 1982) computer based model. These theories have tended to concentrate on the problem of how analogical inferences are made, but they do not provide clearly plausible models of how people make the initial match between the base and target structures (to use Gentner s (1983) terminology) Gentner, for ....

Winston, P. H. (1982). Learning new principles from precedents and exercises. Artificial Intelligence, 19(3), 321-350.

No context found.

P.H. Winston. Learning new principles from precedents and exercises. Artificial Intelligence, 19(3):321--350, 1982.

No context found.

P.H. Winston, "Learning New Principles from Precedents and Exercises," Artificial Intelligence, vol. 19, no. 3, 1982.

No context found.

P.H. Winston, "Learning New Principles from Precedents and Exercises," Artifi- cial Intelligence, vol. 19, no. 3, 1982.

No context found.

P.H. Winston, "Learning New Principles from Precedents and Exercises," Ar- tificial Intelligence, vol. 19, no. 3, 1982.

No context found.

P.H. Winston, "Learning New Principles from Precedents and Exercises," Artifi- cial Intelligence, vol. 19, no. 3, 1982.

No context found.

P.H. Winston, "Learning New Principles from Precedents and Exercises," Ar- tificial Intelligence, vol. 19, no. 3, 1982.

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