Browsing by Author "Maulsby, David L."

Now showing 1 - 10 of 10
  • Thumbnail Image
    ItemOpen Access
    ACQUIRING GRAPHICAL KNOW-HOW: AN APPRENTICESHIP MODEL
    (1988-03-01) Maulsby, David L.; Witten, Ian H.
    This paper studies the acquisition of procedural knowledge, or "know-how", from end-users in the domain of interactive graphics. In order to develop an open-ended system that is not restricted to any particular class of drawings, heavy emphasis is placed on the user interface. Experts (we call them simply "teachers") express procedures constructively, using any of the tools available in the interactive drawing environment. Well-structured procedures, including branches and loops, are inferred using a variety of weak generalization heuristics. The teacher's attention is concentrated on the system's perceptual and inferential shortcomings through a metaphorical apprentice called "Meta-Mouse". Its sensors are predominantly tactile, which forces teachers to make their constructions explicit. Meta-Mouse generalizes action sequences on the fly and eagerly carries out any actions it can predict. Theoretical support for the design comes from two sources: geometric phenomenology, which confirms that powerful problem-solving methods are associated with common-place spatial reasoning; and the fact that Meta-Mouse automatically imposes important "felicity conditions" on the teacher's demonstrations.
  • Thumbnail Image
    ItemOpen Access
    CONSTRAINT-SOLVING IN INTERACTIVE GRAPHICS A USER-FRIENDLY APPROACH
    (1988-11-01) Maulsby, David L.; Kittlitz, Kenneth A.; Witten, Ian H.
    Solving constraints is an important part of interactive graphics, and a number of constraint solvers that operate in this domain have been designed and implemented. However, most of these systems are deficient in two respects: the method of specifying constraints is counter-intuitive, and only a restricted class of constraints is representable. After describing the problems inherent in current systems, we propose a simple constraint solver and its user interface, Metamouse. The user of this system specifies constraints by giving examples in the form of execution traces; the system induces a generalized procedure. Thus constraint specification is natural-the user simply performs his task as usual-and the class of representable constraints includes anything the user could accomplish manually with the graphics editor.
  • Thumbnail Image
    ItemOpen Access
    EVALUATING PROGRAMS FORMED BY EXAMPLE: AN INFORMATIONAL HEURISTIC
    (1990-11-01) Witten, Ian H.; Maulsby, David L.
    The construction of a procedure from a few examples of its execution--a major requirement for practical programming-by-example--is inevitably a drastically underdetermined problem. At the core of any incremental programming-by-example scheme is a heuristic module that creates and modifies the program, or "model," as it is formed. In general there are countless different ways that the model might plausibly be modified; the problem is to prune the set of candidate models to keep it down to a reasonable size. We develop a principled method for evaluating and comparing alternative models of a sequence of actions. Models are finite-state automata and therefore can contain branches and loops. Based on information theory, the method computes the entropy of a model in conjunction with the entropy of the sequence used to form it--a novel form of the "minimum description length" principle. The idea is to measure a model's predictive power, taking into account the extent to which it is justified by the sequence that has been used to create it. The performance of the measure is illustrated on test cases and accords with intuition about when sufficient evidence has accumulated to prefer a more complex model to a simpler one.
  • Thumbnail Image
    ItemOpen Access
    INDUCING PROGRAMS IN A DIRECT-MANIPULATION ENVIRONMENT
    (1988-09-01) Maulsby, David L.; Witten, Ian H.
    End users who need to program within highly interactive direct- manipulation interfaces should be able to communicate their intentions through concrete demonstration rather than in terms of symbolic abstractions. This paper describes a system that learns procedures in interactive graphics taught to it "by example" by minimally trained users. It shows how techniques of machine learning and reactive interfaces can support one another-the former providing generalization heuristics to identify constraints implicit in user actions, the latter offering immediate feedback to help the user clarify hidden constraints and correct errors before they are planted into the procedure. The teacher's attention is focused on the learning system's perceptual and inferential shortcomings through a metaphorical apprentice called Metamouse, which generalizes action sequences on the fly and eagerly carries out any actions it can predict. The success of the induction process is assessed quantitatively by counting erroneous predictions made during example tasks.
  • Thumbnail Image
    ItemOpen Access
    INFERRING GRAPHICAL PROCEDURES: THE COMPLEAT METAMOUSE
    (1990-05-01) Maulsby, David L.; Witten, Ian H.; Kittlitz, Kenneth A.; Franceschin, Valerio G.
    Metamouse is a demonstrational interface for graphical editing tasks within a drawing program. The user specifies a procedure by performing an example execution trace, manipulating objects directly on the screen and creating graphical tools where necessary to help make constraints explicit. The system generalizes the user's action sequence, identifying key features of individual steps and disregarding coincidental events. It creates a program with loops and conditional branches as appropriate, and predicts upcoming actions, thereby reducing the tedium of repetitive and precise graphical editing. It uses default reasoning about graphical constraints to make initial generalizations, and enables the user to correct these hypotheses either by rejecting its predictions or by editing iconic descriptors which it displays after each action.
  • Thumbnail Image
    ItemOpen Access
    METAMOUSE ON TRIAL: CONFESSIONS OF A WANTON TURTLE
    (1991-09-01) Maulsby, David L.; Witten, Ian H.
    We conducted a usability study of Metamouse, a demonstrational interface to a graphics editor that infers complex constraints in a procedural paradigm using graphical construction. Aspects of its inference mechanism and metaphor were tested by having a variety of users perform standard tasks with and without its assistance. We found that people learn to use static rather than dynamic constructions, and the system fails to learn some task decompositions. In particular, its rules for inferring iteration over a set of objects are both inadequate and inadequately disclosed by the metaphor. To address the problem, we propose an explicit representation of sets and generalization over multiple examples.
  • Thumbnail Image
    ItemOpen Access
    MODELING SEQUENCES USING GRAMMARS AND AUTOMATA
    (1994-02-01) Nevill-Manning, Craig G.; Witten, Ian H.; Maulsby, David L.
    Inference of structure from a sequence is useful for explanation, prediction, and compression. One technique described here infers a grammar from a sequence, and presents plausible explanations of how a sequence is structured. It has the added advantage of producing small explanations, and performs extremely well as a data compression technique. A second technique infers an automaton from a sequence, identifying branches, loops, recursive and non-recursive procedures. The two techniques have complementary strengths and weakness, and an inference problem which stymies each technique individually is shown to be amenable to a combination of the two.
  • Thumbnail Image
    ItemOpen Access
    OF MICE AND PENS: HUMAN PERFORMANCE IN DRAWING
    (1988-09-01) Chow, Una Y.; Maulsby, David L.; Witten, Ian H.
    When asked to draw with pen on paper, people exhibit surprising regularity in the apparently free choices they make to execute primitive strokes. Some patterns can be explained in terms of the mechanics of holding the writing instrument; others stem from economy of motion; yet others signify preferred ways of achieving precision when anchoring lines. This paper describes a series of experiments designed to test the extent to which the effects carry over to drawing with mouse and drafting program. It concludes that some habits transfer, albeit in weaker form, despite the fact that mechanical constraints are radically different.
  • Thumbnail Image
    ItemOpen Access
    PROGRAMMING BY EXAMPLE: THE HUMAN FACE OF AI
    (1991-04-01) Witten, Ian H.; MacDonald, Bruce A.; Maulsby, David L.; Heise, Rosanna
    Interactive computer users often find themselves repeatedly performing similar tasks that could be acquired automatically from a teacher. This paper presents principles derived from experience in creating four prototype learners: for technical drawing, text editing, office tasks, and robot assembly. A teaching metaphor (a) enables the user to demonstrate a task by performing it manually, (b) helps to explain the learner's limited capabilities in terms of a persona, and (c) allows users to attribute intentionality. Tasks are represented procedurally, and augmented with constraints. Suitable mechanisms for attention focusing are necessary in order to control inductive search. Hidden features of a task should be made explicit so that the learner need not entertain, and search, all possible missing steps. Key features of the interaction are formalized as "felicity conditions" that help a learner by guaranteeing more explicit, consistent information in demonstrations. Systems that are programmed by human instruction can capitalize on appropriate interactive methods to boost the computational limitations of inductive inference.
  • Thumbnail Image
    ItemOpen Access
    TEACHING A MOUSE HOW TO DRAW
    (1988-01-01) Witten, Ian H.; Maulsby, David L.
    A system is described for programming by example, graphically, which enables untrained end-users to add composite operations to a drawing program using constructive methods traditionally employed in drafting. A pilot experiment showed that considerable extraneous activity occurs in naturally-produced traces. To combat this, full advantage is taken of the interactive situation to constrain induction by suppressing, or at least controlling, variability. A Flatland device called "Meta-Mouse" serves to concentrate the user's attention on the job of teaching a student with limited capabilities. It predicts actions, asks for constructions, solicits input parameters when required, and induces a program (including conditionals and loops). Its behaviors force and help the teacher to satisfy appropriate felicity conditions. Implications for machine learning include the benefits of simulating a pupil to complete the teaching metaphor, and the positive role that close user interaction can play in constraining the search for apt generalizations.