Daniel C. Halbert. SmallStar: Programming by Demonstration in the Desktop Metaphor. In Allen Cypher, editor, Watch What I Do: Programming by Demonstration, pages 102-123. MIT Press, Cambridge, MA, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....process is guided by knowledge that indicates which parts of an example should be generalized, and how to generalize those parts. In this paper, we call the combination of the example provided by the user and the generalization knowledge about the example an annotated example. In some systems [21, 11, 12], the user is responsible for generalizing the example. This simplifies the system, but can be very difficult on the user. At the other end of the spectrum are systems that use artificial intelligence techniques to make inference based on multiple examples [15, 6, 16, 9] possibly without user ....

D. Halbert. SmallStar: Programming by Demonstration in the Desktop Metaphor. In Cypher [7], pages 102--123.

....algebra framework described in this article combines a more expressive and flexible bias with a more natural and straightforward domain representation. Early work that formalized PBD as learning functions to transform application state or data includes Andreae s NODDY [2] and Halbert s SmallStar [12]. Lieberman s Tatlin system [28] infers user actions from observed application state changes. Tatlin learns how to cut and paste data from a calendar application to rows in a spreadsheet. Lieberman also discusses the problem of constructing appropriate data descriptions for the objects manipulated ....

Halbert, D. C.: 1993, `SmallStar: Programming by Demonstration in the Desktop Metaphor'. In: A. Cypher (ed.): Watch What I Do: Programming by Demonstration. Cambridge, MA: MIT Press, pp. 102--123.

....appropriate data descriptions for the objects manipulated by the PBD system, such as the window the user clicked on or the window named Mail ; perhaps the most important aspect of SMARTedit s learning component is its ability to infer the correct data description from examples. See Halbert [7] for additional early work on this topic. SMARTedit is based on the premise that the user is deeply involved in the process of teaching the system how to perform a task, and willing to supervise the learning and execution of the system. In contrast, Ruvini and Dony s work on APE [29] uses version ....

Daniel C. Halbert. SmallStar: Programming by Demonstration in the Desktop Metaphor. In Allen Cypher, editor, Watch What I Do: Programming by Demonstration, pages 102-123. MIT Press, Cambridge, MA, 1993.

....the look of an agent. Parameters, such as the crossing gate in Figure 2, can be references to application world objects. The ability to have application world objects appear in the programming world as they do in the application world significantly helps end users to map between the two worlds [12]. 2.4. Moving between Worlds Tactile programming relies on the ability of end users to easily cross the boundaries between the application world, the programming world and the collaboration world. The philosophy behind tactile programming is not to think of the integration of the collaboration ....

Halbert, D. C., "SmallStar: Programming by Demonstration in the Desktop Metaphor," in Watch What I Do: 8

....novices to learn and use. However, most systems lack the ability for users to access the applications required in their line of work, and the ability to create meaningful data structures that facilitate the organizing and processing of their tasks. Programmable visual shells, such as SmallStar [2], IShell [1] and PURSUIT [7] provide the user with a direct manipulation interface to the file system and allow the creation of programs through interaction with that system. Users can demonstrate programs that copy, delete, compress, and sort files, but including user applications in visual ....

D. C. Halbert, "SmallStar: Programming by Demonstration in the Desktop Metaphor," in A. Cypher (ed.), Watch What I Do: Programming by Demonstration, MIT Press, 1993.

....of heuristics. It also affords an opportunity to assess Cima s performance on real user interactions even before the system is ready for field testing. 2 Data descriptions and utility criteria To model tasks, an agent needs to learn about data, actions, and when to act. Data descriptions (Halbert, 1993) specify criteria for selecting objects, and the results of actions. For instance, suppose the user wants an agent to store email messages from Pattie Maes in the folder Mail from pattie, as shown at the top of Figure 1. The data description sender s id begins pattie tells it which messages to ....

Halbert, D.C. 1993. SmallStar: programming by demonstration in the desktop metaphor. In Cypher (1993a), pp. 103--123.

....of heuristics. It also affords an opportunity to assess Cima s performance on real user interactions even before the system is ready for field testing. 2 Data descriptions and utility criteria To model tasks, an agent needs to learn about data, actions, and when to act. Datadescriptions [Halbert 93] specify criteria for selecting objects, and the results of actions. For instance, suppose the user wants an agent to store email messages from Pattie Maes in the folder Mail from pattie, as shown at the top of Figure 1. The data description sender s id begins pattie tells it which messages ....

Halbert, D.C. "SmallStar: programming by demonstration in the desktop metaphor," in [Cypher 93a], pp. 103--123.

....is an abstract description of a region of interest on a page, in terms of HTML markup elements and text. For instance, the pattern Bulleted List after Heading containing Headlines might describe how news headlines are presented by a particular Web site. Like Halbert s data descriptions [7], patterns are used to describe the parameters of every command in a script. For the copy and paste commands used to build the customized newspaper, both the copied region and the paste location are described by patterns. Patterns are the essential abstractions that allow Turquoise scripts to be ....

....demonstrating different kinds of scripts. The user may explicitly indicate which kind of script is being demonstrated, or Turquoise may use some heuristics to guess which knowledge base is appropriate. 7. RELATED WORK Turquoise is similar to other programming by demonstration systems. SmallStar [7], a system for demonstrating macros in Xerox Star, introduced the notion of data descriptions, which are like Turquoise patterns. TELS [16] infers programs for text editing tasks, such as reformatting a bibliography. Like Turquoise, it abstracts the user s actions into abstract commands like ....

Halbert, D. SmallStar: programming by demonstration in the desktop metaphor. in Watch What I Do: Programming By Demonstration, Allen Cypher, ed. MIT Press, Cambridge MA, 1993, pp. 104-123.

....generated by a PBE system , can be decomposed into the following three sub problems. Problem in PBE (1) Most existing PBE systems represent generated programs in system specific textual forms, which makes it difficult for programmers to understand the programs afterwards. For example, SmallStar[4] represents generated programs in a script form. Problem in PBE (2) It is impossible to perfectly infer the human intention from examples. Thus, PBE systems should offer programmers some means of revising the generated programs interactively. As described above, many systems represent programs ....

....intention from examples. Thus, PBE systems should offer programmers some means of revising the generated programs interactively. As described above, many systems represent programs textually, and as a result, force programmers edit the textual representation of the programs. Examples are SmallStar[4] and AIDE[11] Problem in PBE (3) Most importantly, programmers often have difficulties in deciding what examples they should provide to PBE systems. For example, in DEMO II[2] programmers themselves are responsible for deciding what sequence of examples to give, which is often more difficult ....

Halbert, D. C., "SmallStar: Programming by Demonstration in the Desktop Metaphor," in Watch What I Do: Programming by Demonstration (A. Cypher, ed.), ch. 5, pp. 103--124, MIT Press, 1993.

....as Undo, Help, and Task Analysis. 5 Related Work For people interested in learning more about Programming by Demonstration, 4] presents a thorough overview and history of the field and describes the current state of the art. The crucial problem of determining user intent was first described in [9]. While many systems have made inroads on this problem, perhaps the most promising is Cima [17] a learning architecture being developed specifically for PBD systems. We are currently pursuing ways to integrate our work with the Cima environment. While there are many user interface specification ....

Daniel C. Halbert. SmallStar: Programming by demonstration in the desktop metaphor. In Allen Cypher, editor, Watch What I Do: Programming by Demonstration, pages 103--124. MIT Press, Cambridge, MA, 1993.

....1987) by requiring that the learned description not only classify correctly but also specify all features of data required for a given type of action, and include all features suggested by the user. 2 To model tasks, an agent needs to learn about data, actions, and when to act. Data descriptions (Halbert, 1993) specify criteria for selecting objects, and the results of actions. Conventional machine learning algorithms learn to classify examples. But agents do things with data, and to be useful, data descriptions may require features in addition to those needed for classification. This is one reason why ....

D. C. Halbert (1993) "SmallStar: programming by demonstration in the desktop metaphor," in Cypher (1993a), pp. 103--123.

....box. After generalizing, the value displayed in the script is replaced with a descriptive placeholder. Bringing up the dialog box for a placeholder will allow it to be edited, or changed back into a constant. This is related to the data description property sheets in the SmallStar visual shell [1], but here they are domain independent. If an application has a special type of value, the programmer can add a new dialog box to handle it, but the built in dialog boxes seem sufficient for many applications. Also, the dialog boxes have a custom option that will in the future link to a ....

....which has not been done previously. Vmacs [5] was an early attempt to apply Emacs ideas to graphics, but it did not support any scripting by example, and instead concentrated on parsing of freehand drawings. The idea of dialog boxes to generalize parameters in scripts was used in the SmallStar [1] visual shell, and Topaz provides this capability for any graphical application. Graphical search was introduced in [4] for a graphics editor, and Topaz allows it to be used in scripts and for multiple types of applications. Topaz is also the first to allow the generalization of the parameters of ....

Halbert, D.C. "SmallStar: Programming by Demonstration in the Desktop Metaphor," in Watch What I Do: Programming by Demonstration. 1993. Cambridge, MA: MIT Press. pp. 102-123.

....as Undo, Help, and Task Analysis. 5 Related Work For people interested in learning more about Programming by Demonstration, 4] presents a thorough overview and history of the field and describes the current state of the art. The crucial problem of determining user intent was first described in [9]. While many systems have made inroads on this problem, perhaps the most promising is Cima [17] a learning architecture being developed specifically for PBD systems. We are currently pursuing ways to integrate our work with the Cima environment. While there are many user interface specification ....

Daniel C. Halbert. SmallStar: Programming by demonstration in the desktop metaphor. In Allen Cypher, editor, Watch What I Do: Programming by Demonstration, pages 103--124. MIT Press, Cambridge, MA, 1993.

No context found.

Daniel C. Halbert. SmallStar: Programming by Demonstration in the Desktop Metaphor. In Allen Cypher, editor, Watch What I Do: Programming by Demonstration, pages 102-123. MIT Press, Cambridge, MA, 1993.

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