P. Langley, H. A. Simon, G. L. Bradshaw, and J. M. Zytkow, Scientific Discovery: Computational Explorations of the Creative Processes. Cambridge, MA: MIT Press, 1987.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....type 2 discovery mechanisms and derive complexity measures directly from their performance statistics. Unfortunately, the choice of candidate mechanisms is much smaller in the type 2 case than it is in the type 1 case. If we leave aside those mechanisms which are based on domain specific search [21, 22, 23, 24], we are left with a fairly small field, the main contenders in which appear to be the well known connectionist learning algorithms such as backpropagation [1] and cascade correlation [25] However, the jury is still very much out on the degree to which such algorithms are capable of discovering ....

Langley, P., Simon, H., Bradshaw, G. and Zytkow, J. (1987). Scientific Discovery: Computational Explorations of the Creative Processes. Cambridge, Mass.: MIT Press.

....each of these stages, but none has automated all of them. Statistical packages provide facilities for hypothesis testing, but must rely on the user to formulate hypotheses, select the appropriate analysis, prepare the data, and explain the results. Intelligentdiscovery systems, such as Bacon [6], Tetrad [4] and C4.5 [8] formulate hypotheses of a specific form and then perform analysis in this context, but rarely have facilities for gathering and preparing data or explaining the results of analysis. Other discovery systems have addressed these latter concerns in the context of certain ....

Pat Langley, Herbert A. Simon, Gary L. Bradshaw, and Jan M. Zytkow. Scientific Discovery: Computational Explorations of the Creative Processes. MIT Press, Cambridge, MA, 1987.

....cases, it is not necessary to verify all three conditions. For example, if the statement Y =3X 5were derived from the method of least squares, the residuals ffl are independentofX by definition. Increasing the exponent under these conditions is essentially the heuristic introduced by bacon.3 [8]. 2.3 Automated Experimentation One of sea s important features is its ability to design and run experiments with a simulator. For our purposes, the simulator can be any LISP based computer program. Sea generates an experiment plan for use by the instrumentation package clip [2] which in turn ....

....these decisions will become more frequentand mayoverburden the user. Thus, weareinterested in developing policies for automatic focusing and refocusing. Manysuch policies can be hand coded from successful scientific strategies, such as the scientific discovery heuristics of programs like Bacon [8] and IDS [9] 3.2 Structured Domain Knowledge Knowledge about the scientific domain in question is acquired from the user on an as needed basis. This knowledge is stored as collections of statements called views. Aviewisasetof statements with a consistentinterpretation of simulation behavior. ....

Pat Langley, Herbert A. Simon, Gary L. Bradshaw, and Jan M. Zytkow. Scientific Discovery: Computational Explorations of the Creative Processes. MIT Press, Cambridge, MA, 1987.

....candidate for a successful synthesis. MECHEM does not currently have databases of any kind, with the exception of a list of species free energies that was obtained from a commercial program called ASPEN. Another body of work related to MECHEM consists of the programs reported in Langley et al. [17], although the goal of their work was to find simple mechanisms able to reconstruct specific historical discoveries with modest amounts of computation. Three of the four programs they describe (GLAUBER, STAHL, and DALTON) deal with chemical reactions. In terms of the programs methods, one ....

P. Langley, H.A. Simon, G.L. Bradshaw, and J.M. Zytkow. Scientific Discovery: Computational Explorations of the Creative Processes. MIT Press, Cambridge, Mass., 1987.

....discovery is concerned with reconstructing (or constructing de novo) the logic, psychology, or history of scientific discovery by means of computation. Examples of work in this area include more or less psychological reconstructions of historical discoveries [Karp, 1993, Kulkarni and Simon, 1988, Langley et al. 1987, Ledesma et al. 1994, Ledesma et al. 1993] analysis of the computational and heuristic logic of some discovery task [Fischer and Zytkow, 1990, Lindsay et al. 1980, Valdes, 1992, Valdes, in press] conceptual generalizations of distinct discovery systems [Langley and Zytkow, 1989, Valdes et ....

....to the same chunk. Figure 5 reveals a pronounced tendency to place pieces successively in the same direction (0 ffi turning angle) although not necessarily horizontally. Once again, this pattern had not been noticed by the experimenters. 5 DISCUSSION One lesson from the work on BACON [Langley et al. 1987] and its successors is that a single heuristic can by itself be quite powerful in making significant discoveries in scientific data. The concept of data driven discovery describes such cases, in which little theoretical knowledge is employed; most of the power comes from the data and from the ....

[Article contains additional citation context not shown here]

Langley, P., Simon, H., Bradshaw, G., and Zytkow, J. (1987). Scientific Discovery: Computational Explorations of the Creative Processes. MIT Press, Cambridge, Mass.

.... [13, 17, 18, 22] and just recently, it became the research subject of some cognitive scientists and computer scientists who believe that the process of a scientific discovery can be described and modeled in a normal way and therefore it can be simulated by computer programs automatically [15, 21, 23]. First of all, we present here some significant fundamental observations and assumptions, which underlie our research direction, on scientific discovery processes and their automation as follows: 1) New conditionals are epistemic goals of any scientific discovery: Any scientific discovery ....

P. Langley, H. A. Simon, G. L. Bradshaw, and J. M. Zytkow, "Scientific Discovery - Computational Explorations of the Creative Processes," MIT Press, 1987.

....and compare the designs of discovery programs that are intended to be used as collaborators by scientists. 1 1 Introduction Early work on machine scientific discovery such as Logic Theorist [25] DENDRAL [23] and AM [22] and later work on the cognitive modelling of historical discovery (e.g. [21]) have shown that heuristic search in combinatorial spaces is a useful, general framework for automating and explaining discovery. However, it has been unclear what further generality could be found among programs that accomplish diverse tasks in different sciences. Absent general principles, ....

....of programs may be conceded from the start. Other programs have done science well (e.g. 33, 24, 30, 44, 47, 48] but this list suffices for our purposes. There are also exploratory programs that have illuminated aspects of scientific reasoning without yet serving as collaborators (e.g. [21, 4, 7, 1, 5, 36, 53, 18, 22]) these are outside the scope of our analysis. Each program handles the four dimensions in multiple ways, but we will usually mention only one way for each program dimension combination. Our analysis will be a posteriori: we examine systems that are successful as determined by publication in the ....

P. Langley, H. Simon, G. Bradshaw, and J. Zytkow. Scientific Discovery: Computational Explorations of the Creative Processes. MIT Press, Cambridge, Mass., 1987.

.... discovery in databases [HCC92] Lub89] PS91b] Related, but not directly applicable, work includes the induction of classification rules [BFOS84] Cat91] FWD93] HCC92] Qui93] discovery of causal rules [CH92] Pea92] learning of logical definitions [MF92] Qui90] fitting of functions to data [LSBZ87] Sch90] and clustering [ANB92] C 88] Fis87] The closest work in the machine learning literature is the KID3 algorithm presented in [PS91a] If used for finding all association rules, this algorithm will make as many passes over the data as the number of combinations of items in the ....

P. Langley, H. Simon, G. Bradshaw, and J. Zytkow. Scientific Discovery: Computational Explorations of the Creative Process. MIT Press, 1987.

....functions [ Minsky and Papert, 1969, Nilsson, 1965 ] 3 Case studies The process of scientific discovery begins with an exploratory phase in which a phenomenon of interest is observed. Based on these observations, hypotheses are formulated and experiments are designed to test these hypotheses [ Langley et al. 1987 ] In this section, we describe how the hyperplane animator served as an important tool in this initial phase of exploration of the dynamics of weight changes in backpropagation networks, and how the resulting experiments have led to important new insights and algorithms. 3.1 Weight magnitudes, ....

Pat Langley, Herbert A. Simon, Gary L. Bradshaw, and Jan M. Zytkow. Scientific Discovery: Computational Explorations of the Creative Process. MIT Press, Cambridge, Massachusetts, 1987.

....discussion to programs whose main objective is to invent concept definitions and make conjectures in pure mathematics. This leaves out automated theorem provers (which discover proofs) and programs which discover mathematical results in other domains, such as the very important BACON programs, Langley et al. 1987). To compare and contrast the discovery programs, we detail what the project aims were, how the program worked and what contributions the programs made to mathematics and the understanding of mathematical discovery. We pay particular attention to the measures employed to estimate how interesting a ....

P Langley, H Simon, G Bradshaw, and J Zytkow. Scientific Discovery - Computational Explorations of the Creative Processes. MIT Press, 1987.

....approaches to constructing nominal and continuousvalued attributes. Bacon and Induce also construct new continuous valued attributes. On the other hand, Induce, AQ17 dci, and AQ17 mci construct attribute counting attributes that have ordered discrete values. The scientific discovery system Bacon [Langley et al. 1987a; 1987b] discovers numeric laws. It constructs new attributes (called terms) such as X Theta Y , X=Y , X N , log(X) and sin(X) by using mathematical operators, where X and Y are primitive continuous valued attributes (variables) Given a set of observations with each represented as the values ....

.... of a line, e.g. y 2 Gamma y 1 ) x 2 Gamma x 1 ) In addition, to deal with more complex relations between X and Y , Bacon considers transformations of both the independent and the dependent terms, such as inverse(Y ) sin(Y ) and log(Y ) using a simple minded generate and test strategy 50 [Langley et al. 1987a] To deal with noise, it requires a near constant differential instead of a strict constant to be found. The rule learning algorithm Induce [Larson and Michalski, 1977; Michalski, 1980a; 1983] also uses mathematical operators as constructive operators. It constructs new attributes (descriptors) ....

[Article contains additional citation context not shown here]

P. Langley, H.A. Simon, G.L. Bradshaw, and J.M. Zytkow, Scientific Discovery: Computational Explorations of the Creative Processes, Cambridge, MA: MIT Press.

....for combining breaking apart such substances, structures and sub structures Previous work in computational scientific discovery has addressed how each of these types of chemical knowledge can be used to elucidate the others. For example, the STAHL and DALTON systems (Zytkow Simon, 1986; Langley et al. 1987) use the concept of chemical reaction to determine the components of a substance (STAHL) and its molecular composition (DALTON) Related systems are REVOLVER (Rose Langley, 1986, 1988; Rose, 1989) BR3 (Kocabas, 1991) and GELL MANN (Fischer Zytkow, 1990, 1992) These systems all use a ....

....Informal Qualitative Models, or IQMs (Sleeman et al. 1989) They are informal and qualitative in the sense that they cannot be directly verified by observation. For example, they cannot be verified by the presence or absence of particular reactions. A critique of the BACON family of programs (Langley et al. 1987) led us to articulate the concept of IQMs. Besides being concerned about noise handling capabilities in BACON, we were exercised by the fact that such quantitative law discovery systems needed to be given the dependent and independent variables, and their associated sets of values. Sleeman et al. ....

Langley, P., Simon, H.A., Bradshaw, G.L. and Zytkow, J.M. (1987) Scientific Discovery: Computational Explorations of the Creative Processes , Cambridge, MA: MIT Press.

....discussion to programs whose main objective is to invent concept definitions and make conjectures in pure mathematics. This leaves out automated theorem provers (which discover proofs) and programs which discover mathematical results in other domains, such as the very important BACON programs, [19]. To compare and contrast the discovery programs, we detail what the aims of the project were, how the program worked and what contributions the programs made to mathematics and the understanding of mathematical discovery. We pay particular attention to the measures employed to estimate how ....

P Langley, H Simon, G Bradshaw, and J Zytkow. Scientific Discovery - Computational Explorations of the Creative Processes. MIT Press, 1987.

....supervisor or a teacher. Decision tree methods [15, 11] and neural network learning [6] fall into this category. In unsupervised learning, the agent is given a collection of input values but no output values. The agent has to find regularity in the inputs by itself. Clustering [5] and discovery [8] methods fall into this category. Unsupervised learning methods are widely applied under the popular name data mining nowadays. Reinforcement learning [19] falls in between supervised and unsupervised learning. In reinforcement learning, an agent does not receive input output examples from the ....

Pat Langley, Herbert Simon, Gary L. Bradshaw, and J. M. Zytkow. Scientific Discovery: Computational Explorations of the Creative Processes. MIT Press, Cambridge, Massachusetts, 1987.

....Lenat [3] built the AM system, which, given basic knowledge about a domain, was able to construct interesting new concepts in that domain, and this was followed by the EURISKO system that had capabilities for extending its repertory of discovery heuristics. Langley, Simon, Zytkow and Bradshaw [2] devised a series of programs which could derive scientific laws from data (BACON, GLAUBER, STAHL, DALTON, and others) which in turn suggested a theory of human scientific discovery. As noted earlier, efforts at modelling discovery processes have sometimes been aimed at developing a theory of ....

P. Langley, H. Simon, G. Bradshaw, and J. Zytkow. Scientific Discovery: Computational Explorations of the Creative Processes. MIT Press, Cambridge, Mass., 1987.

.... of Computing Science, University of Aberdeen The discovery of scientific knowledge is one of the most challenging tasks that confront humans, yet cognitive science has made considerable progress toward explaining this activity in terms of familiar cognitive processes like heuristic search (e.g. Langley et al. 1987). A main research theme relies on selecting historical discoveries from some discipline, identifying data and knowledge available at the time, and implementing a computer program that models the processes that led to the scientists insights. The literature on computational scientific discovery ....

Langley, P., Simon, H. A., Bradshaw, G. L., & Zytkow, J. M. (1987). Scientific discovery: Computational explorations of the creative processes .

....Discovery Our research on computational scientific discovery draws on two previous lines of work. One approach, which has an extended history within artificial intelligence, addresses the discovery of explicit quantitative laws. Early systems for numeric law discovery like Bacon (Langley, 1979; Langley et al. 1987) carried out a heuristic search through a space of new terms and simple equations. Numerous successors like Fahrenheit ( Zytkow et al. 1990) and RF5 (Saito Nakano, 1997) incorporate more sophisticated and more extensive search through a larger space of numeric equations. 10 The most relevant ....

Langley, P., Simon, H. A., Bradshaw, G. L., & Zytkow, J. M. (1987). Scientific discovery: Computational explorations of the creative processes . Cambridge, MA: MIT Press.

....conjectures. Further model improvements must take into account properties such as static and dynamic friction. In our example, we merely add a known variable (angular velocity) to the model. Other work on automation of scientific discovery proposed various reasons for postulating new variables (Langley et. at, 1987; Kocabas, 1991; Valdes Perez, 1993) Modeler s knowledge of hidden structure of the phenomenon O grows until the model satisfies the verification criteria at all levels, or an impass is reached and modeling must stop until more background knowledge are gained or new observations are made of the ....

Langley, P., Simon, H. A., Bradshaw, G. L. & Zytkow, J. M. 1987. Scientific discovery: Computational explorations of the creative processes. Cambridge, MA: MIT Press.

....assumed that the vote count is similar to the natural phenomena. Therefore, attribute oriented induction with data sampling will be effective for rule acquisition in a dynamic environment. Discovery of functional relationships in numerical data has been studied in the programs such as Bacon[8]. Generalization (or abstraction) is also an essential technique in such programs to grasp the knowledge about the status of a complex system. For numerical values, aggregation of variables is based on the eigenvalues of the system matrix [5] On the other hand, in our proposed induction ....

P. Langley, H.A. Simon, G.L. Bradshaw and J.M. Zytkow, "Scientific Discovery: Computational Explorations of the Creative Processes. MIT Press, 1987.

....restricted the type of concept formed to definitions, loosely speaking, the sentences which appear under the definition heading of finite group theory texts. As detailed later, further uniformity is achieved by thinking of each definition as a function mapping a group to some output. 2 See [Langley et al. 87] 3 See [Sims 90] A major goal in the development of finite group theory is the classification of finite groups. For instance, there is a classification of finite Abelian groups which can be found in most elementary group theory texts. More remarkably, in 1980, a complete classification of ....

P Langley, H A Simon, G L Bradshaw, and J M Zytkow. Scientific Discovery - Computational Explorations of the Creative Processes. MIT Press, 1987.

No context found.

P. Langley, H. A. Simon, G. L. Bradshaw, and J. M. Zytkow, Scientific Discovery: Computational Explorations of the Creative Processes. Cambridge, MA: MIT Press, 1987.

No context found.

Langley, P., Simon, H., Bradshaw, G. and Zytkow, J. (1987). Scientific Discovery: Computational Explorations of the Creative Processes. Cambridge, Mass.: MIT Press.

No context found.

P. Langley, H. A. Simon, G. L. Bradshaw, and J. M. Zytkow, Scientific Discovery: Computational Explorations of the Creative Processes. Cambridge, MA: MIT Press, 1987.

No context found.

LANGLEY P. et al., Scientific Discovery: Computational Explorations of the Creative Process, The MIT Press, Cambridge, Mass. 1987.

No context found.

Langley, P., Simon, H., Bradshaw, G., and Zytkow, J., 1987. Scientific Discovery: Computational Explorations of the Creative Process. MIT Press, Cambridge, MA.

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