D.B. Lenat and J.S. Brown. Why am and eurisko appear to work. Artificial Intelligence, 23:269--294, 1983.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....much of the generated solution is due to the algorithm and how much to the background knowledge. In particular, we want to be able to decide whether the solution provided by the learner is implicitly contained in the background knowledge since in this case no real learning would have taken place. [2] In the familiar case of concept learning it is conventional to think of the learner as searching through a space of hypotheses for one that satisfies the constraints imposed by the training set [3] In this model, the only way background knowledge can affect the learning is by causing the ....

Lenat, D. and Brown, J. (1984). Why AM and EURISKO appear to work. Artificial Intelligence, 23, No. 3 (pp. 269-294).

....Such a system is not likely to be very useful, however, because it will spend most of its time deriving uninteresting theorems. So the success of this model of continual computation will hinge on whether we can find meaningful criteria for the interestingness of a theorem. In the classical AM [9 11], the system relies largely on human judgment determine interestingness. In a survey of several automated discovery programs, Colton and Bundy [12] identify several properties of concepts which seem to be relevant to their interestingness, such as novelty, surprisingness, understandability, ....

Lenat, D.B., Brown, J.S.: Why AM and EURISKO appear to work. Artificial Intelligence 23 (1984) 269--294

....Such a system is not likely to be very useful, however, because it will spend most of its time deriving uninteresting theorems. So the success of this model of continual computation will hinge on whether we can find meaningful criteria for the interestingness of a theorem. In the classical AM [12 14], the system relies largely on human judgment determine interestingness. In a survey of several automated discovery programs, Colton and Bundy [15] identify several properties of concepts which seem to be relevant to their interestingness, such as novelty, surprisingness, understandability, ....

Lenat, D.B., Brown, J.S.: Why AM and EURISKO appear to work. Artificial Intelligence 23 (1984) 269--294

....induction, unsupervised learning. 1 Introduction Lenat s program AM was the first automatic theory construction system to attract the attention of AI community. Beginning with pre numerical concepts, it rediscovered some well known concepts and conjectures in elementary mathematics [7], such as the prime numbers, the deMorgan s Law and the Goldbach s conjecture. AM s initial success and later inability to generate new results, were analyzed mainly by its creator. AM represented concepts by LISP programs that generated examples of the concept, and relied on the syntactic ....

Douglas B. Lenat. Why AM and EURISKO appear to work. Journal of Artificial Intelligence, 23, 1984.

....rather than available for further consideration. This is not to say that active logic has a built in general purpose concept formation mechanism; but it does have the expressive power to represent and reason with such formations, if they were made available, perhaps along lines of AM (Lenat 1982; Lenat Brown 1984). Furthermore, as seen earlier, active logic allows for recognition that a given statement is already known, or that it s negation is known, or that neither is known, thereby avoiding re derivation of a theorem. Similarly, if such a purported human style theorem prover (HSTP) that is allowed to ....

....not likely to be very useful, however, because it will spend most of its time deriving uninteresting theorems. So the success of this model of continual computation will hinge on whether we can find meaningful criteria for the interestingness of a theorem. In the classical AM (Lenat 1982; 1983; Lenat Brown 1984), the system relies largely on human to provide the judgment of interestingness. In a survey of several automated discovery programs, Colton Bundy 1999) identify several properties of concepts which seem to be relevant to their interestingness, such as novelty, surprisingness, ....

Lenat, D. B., and Brown, J. S. 1984. Why AM and EURISKO appear to work. Artificial Intelligence 23(3):269--294.

....and Theorems When solving problems, mathematicians follow a Mathcycle [VeVe94] conception, naive formulation, exploration, tentative proof, formulation, proof, publication, education, and use. Many packages which aid mathematicians in some of these steps have been developed, e.g. Am [LeBr84] for concept formulation, CAS for application of algorithms, ATP for verification and discovery of theorems, specification languages and knowledge representation. However, few mathematicians use these systems as everyday tools, because of some severe drawbacks which make them hard to use. ....

D.B. Lenat, J.S. Brown, Why AM and EURISKO Appear to Work , Artificial Intelligence 23, pp. 269--294, Elsevier, 1984.

....example of this design problem. Programs are syntactic machines, and if you want meaningful outputs, you have to write programs whose syntactic operations are meaningful to you. Doug Lenat and John Seely Brown recognized this problem in a paper called Why AM and Eurisko Appear to Work [20]. The AM system discovered many concepts in number theory, but when the AM approach was tried in other domains, it didn t work as well. Lenat and Brown concluded that AM worked because syntactic Lisp operations on Lisp representations of mathematical concepts often produced meaningful new ....

Douglas B. Lenat and John S. Brown. Why AM and EURISKO Appear to Work, AAAI83. Pp. 236-240.

....Therefore, if a user expressed an interest in a concept, the theory would develop around that concept. There has been much debate about the AM program. In [31] Hanna and Ritchie were particularly critical of the methods AM used and the accuracy of Lenat s description of his work, and in [23], Lenat replied to this criticism. The main contribution of Lenat s work is an inspiration for how computers could do mathematics, i.e. by creating concepts and conjectures of many different types and using heuristic methods such as analogy and symmetry to explore a domain. 2.2 The GT Program The ....

D Lenat and J Brown. Why AM and EURISKO appear to work. Artificial Intelligence, 23, 1984.

.... heuristic to evaluate new concepts, which are found using heuristic production rules. AM is essentially a best first search with carefully tailored search operators (the heuristic production rules) and evaluation function (the interestingness heuristic) EURISKO [Lenat, 1982a, Lenat, 1982b, Lenat and Brown, 1984] is an extension of AM that adds a heuristic description language, allowing the system to be applied to new domains and to find new heuristics using a meta discovery process. Domain specific information is still required, but the meta rules for finding new heuristics are somewhat more general. ....

Douglas B. Lenat and John Seely Brown. Why AM and EURISKO appear to work. Artificial Intelligence, 23:269--294, 1984.

....model for creativity. Some of these programs are designed to be creative within artistic domains, such as Cohen s drawing program aaron [19] and programs designed to write stories [68, 89] Other such programs work in scienti c and mathematical domains, such as Lenat s controversial AM [60, 90], work on conjecture making in graph theory [30] and various programs which make conjectures about chemical reactions [59] The important feature of these programs is that they are 6 not focussed on solving speci c problems, but they take a large database of information and attempt to induce ....

D. B. Lenat and J. S. Brown. Why am and eurisko appear to work. Articial Intelligence, 23:269{ 294, 1984.

....reason that inductive learning algorithms are not more widely used in problemsolving systems is their sensitivity to the way in which information is represented. Some of the best known successes in machine learning have been due in part to careful or fortunate choices of representations (e.g. AM [Lenat Brown, 1984]) When the same algorithms are applied with less carefully chosen vocabularies, they may fail to learn anything useful. As a result, machine learning techniques are impossible to apply without confronting one of the oldest and most pervasive problems in Artificial Intelligence: how to choose an ....

....Machine learning is a field that has historically been empirical; researchers construct programs, experiment with them, and publish the results. This methodology sometimes makes it difficult to evaluate a learning algorithm, since even its author may not understand why it works (e.g. AM [Lenat Brown, 1984]. As a result, researchers are trying to be more careful about understanding not only how the algorithm works, but why it works. In keeping with this trend in machine learning, the thesis research will occur along two dimensions. First, the hypotheses H1 and H2 will be validated empirically. H2 ....

Lenat, D. B., & Brown, J. S. (1984). Why AM and EURISKO appear to work. Artificial Intelligence, 23, 269-294.

....using a function of the strengths of reasons given for performing them, the interestingness of their concepts, and the interestingness of the type of task being performed. AM used heuristics to perform tasks and to propose new tasks. The reader is directed to (Lenat 1982A) for details of AM and to (Lenat and Brown 1984; Ritchie and Hanna 1984; Haase 1990; Shen 1990) for discussions of AM. AM s framework is general, initially being used in AM to discover quite a few new (to AM) mathematical concepts, such as prime numbers and Goldbach s conjecture. Further work by Lenat (Lenat 1990) demonstrated the framework s ....

Lenat, D. B. and Brown, J. S. 1984. Why AM and Eurisko Appear to Work. Artificial Intelligence 23: 269-294.

....reason that inductive learning algorithms are not more widely used in problem solving systems is their sensitivity to the way in which information is represented. Some of the best known successes in machine learning have been due in part to careful or fortunate choices of representations (e.g. AM [Lenat Brown, 1984]) When the same algorithms are applied with less carefully chosen vocabularies, they may fail to learn anything useful. As a result, the success in using inductive learning algorithms varies from person to person. Constructing a good vocabulary by hand can take a long time, because it is ....

Lenat, D. B., & Brown, J. S. (1984). Why AM and EURISKO appear to work. Artificial Intelligence, 23, 269-294.

....and water and teakettles. The same sort of chicken and egg relationship characterizes CYC and ML because learning occurs at the fringe of what one already knows. Therefore, in the early 1980s, when the rest of the world was so enthusiastic about NLU, ML, and AI in general, we were pessimistic [2]. We concluded the only way out of this codependency would be to prime the pump by manually crafting a million axioms covering an appreciable fraction of the required knowledge. That knowledge would serve as a critical mass, enabling further knowledge collection through NLU and ML, beginning in ....

Lenat, D. B.and Brown, J. S. Why AM and Eurisko appear to work. J. Artif. Intell., 23 (1984), 269--294.

No context found.

D.B. Lenat and J.S. Brown. Why am and eurisko appear to work. Artificial Intelligence, 23:269--294, 1983.

No context found.

D Lenat and J Brown. Why AM and EURISKO appear to work. Artificial Intelligence, 23, 1984.

No context found.

D. Lenat and J. S. Brown. Why AM and EURISKO appear to work. Artificial Intelligence, 23:269--294, 1984.

No context found.

D. Lenat and J. S. Brown. Why AM and EURISKO appear to work. Artificial Intelligence, 23:269--294, 1984.

No context found.

D Lenat and J Brown. Why AM and EURISKO appear to work. Arti cial Intelligence, 23, 1984.

No context found.

D.B. Lenat and J.S. Brown. WhyAM and Eurisko appear to work. Artificial Intelligence, 23(3):269--294, 1984.

No context found.

J. Brown, and D. Lenat, Why AM and EURISKO Appear to Work, Arti cial Intelligence 23 (1984) 269-294.

No context found.

J. Brown, and D. Lenat, Why AM and EURISKO Appear to Work, Arti cial Intelligence 23 (1984) 269-294.

No context found.

D Lenat and J Brown. Why AM and EURISKO appear to work. Artificial Intelligence, 23, 1984.

No context found.

J. Brown, and D. Lenat, Why AM and EURISKO Appear to Work, Arti- cial Intelligence 23 (1984) 269-294.

No context found.

Lenat, D.B., Brown, J.S., Why AM and Eurisko Appear to Work, Proceedings of the Third National Conference on Artificial Intelligence, August 22-26, 1983, Washington, D.C., AAAI, pp. 236-240.

First 50 documents

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