Robert St. AmantandPaul R. Cohen. A planner for exploratory data analysis. In Proceedings of the Third International ConferenceonArtificial IntelligencePlanning Systems, pages 205--212. AAAI Press, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....and actions must be applied selectively to maximize efficiency. SEA uses a partial hierarchical planning mechanism to guide its search through a space of hypotheses [3,4,7] SEA s planner was originally developed for a similar agent, 1 the Assistant for Intelligent Data Exploration (AIDE) [1]. AIDE works with a user to discover the structure of a dataset using exploratory statistical procedures. Many of its plans are relevant to SEA s task, and we expect to eventually integrate the two systems. The planner selects plans from SEA s plan library to match posted empirical goals. The ....

Rob St. Amant. A planner for exploratory data analysis. Submitted to the Third International Conference on Artificial Intelligence Planning Systems (APIS-96).

....relationships between the variables. Statements are initially formulated by the user or through heuristic reasoning. Each statement is evaluated in light of experiment data. SEA uses a partial hierarchical planner developed by Rob St. Amant for his Assistant for Intelligent Data Exploration (AIDE)[1]. The planning environment is partial in that it interleaves planning phases with execution of the plan. This approach is beneficial in domains such as data exploration and hypothesis testing, where the goals of the planner are poorly defined. DRAFT 11 20 95 10 There are many ways to address the ....

Rob St. Amant. A planner for exploratory data analysis. Submitted to the Third International Conference on Artificial Intelligence Planning Systems (APIS-96).

....obtaining that value, and reach a conclusion based on the probability. Plans maycontain iteration, conditionals, executable statements, variable bindings, and subgoals# thus, the planning language is essentially a highlevel programming language. The planner used by SEA was developed by St. Amant[9] and applied to the problem of exploratory data analysis. In addition to the basic PHP mechanisms mentioned above, the planner includes two important features. First, it can be easily extended to accomodate new control constructs in the planning language. More importantly, the planner provides a ....

Robert St. AmantandPaul R. Cohen. A planner for exploratory data analysis. In Proceedings of the Third International ConferenceonArtificial IntelligencePlanning Systems, pages 205--212. AAAI Press, 1996.

....is essentially bottom up, composing elementary actions into a high level plan, PHP is top down, gradually instantiating an abstract plan until an executable action is encountered. Sea uses a planner originally developed for a similar system, the Assistant for Intelligent Data Exploration (aide)[10]. This planner employs a mechanism called focusing to manage decision points in the evolving plan. Whenever a new goal is posted, the planner selects a set of plan fragments that unify with the goal. If the set is empty, the goal cannot be satisfied. If there is only one plan in the set, it is ....

Robert St. Amant and Paul R. Cohen. A planner for exploratory data analysis. In Proceedings of the Third International ConferenceonArtificial Intelligence Planning Systems, pages 205-- 212. AAAI Press, 1996.

....is essentially bottom up, composing elementary actions into a high level plan, PHP is top down, gradually instantiating an abstract plan until an executable action is encountered. Sea uses a planner originally developed for a similar system, the Assistant for Intelligent Data Exploration (aide) [6]. This planner employs a mechanism called focusing to manage decision points in the evolving plan. Whenever a new goal is posted, the planner selects a set of plan fragments that unify with the goal. If the set is empty, the goal cannot be satisfied. If there is only one plan in the set, it is ....

Robert St. Amant and Paul R. Cohen. A planner for exploratory data analysis. In Proceedings of the Third International Conference on AI Planning Systems (AIPS-96), pages 205--212, 1996.

....easily be generated. 3 Navigation and the data analysis process It will be convenient to discuss interaction techniques in the framework of a specific system. Our discussion will center on Aide, an assistant for intelligent data exploration, which we have developed over the past several years [15, 14]. Aide is a knowledge based system that incrementally explores a dataset, guided by user directives and its own evaluation of indications in the data. Aide s knowledge base implements simple strategies for exploratory data analysis (EDA) These strategies are represented as plans, a refinement of ....

Robert St. Amant and Paul R. Cohen. A planner for exploratory data analysis. In Proceedings of the Third International Conference on Artificial Intelligence Planning Systems, pages 205--212. AAAI Press, 1996.

....in user needs; they contain mechanisms for maintaining shared, implicit knowledge. Burstein and McDermott (1996) expand on these and related issues in a summary of the state of the art. The AP planner is a straightforward reactive, scriptbased planner. We have described its design elsewhere (St. Amant Cohen 1996); a brief summary will be enough to support our discussion. A planning session begins with the establishment of a top level goal. AP searches through its library for an appropriate plan and expands it into its component control constructs: sequences, conditionals, iteration, mapping, and so forth. ....

St. Amant, R., and Cohen, P. R. 1996. A planner for exploratory data analysis. In Proceedings of the Third International Conference on Artificial Intelligence Planning Systems, 205--212. AAAI Press.

....results into a coherent larger picture. The system is mixed initiative, autonomously pursuing high and low level goals while still allowing the user to inform or override its decisions. Elsewhere we have described Aide s operations and primitive data structures [22] its planning representation [23], its user interface [25, 24] and the system as a whole [21] This progress report discusses a recent evaluation we conducted with Aide and explains why we believe that this line of research is important to AI and statistics researchers. We will begin with a very brief overview of the system. The ....

....of building and evaluating Aide. We end with a discussion of the generality of our results and the potential for future work. 1 AIDE and Planning Aide s design exploits a striking similarity between interactive data exploration and a type of AI planning known as partial hierarchical planning [22, 23]. Aide maintains a library of over a hundred plans and control rules representing knowledge about how statistical procedures are carried out. Each plan is designed to capture an element of common statistical practice, such as the examination of residuals after fitting a function to a relationship, ....

Robert St. Amant and Paul R. Cohen. A planner for exploratory data analysis. In Proceedings of the Third International Conference on Artificial Intelligence Planning Systems, pages 205-- 212. AAAI Press, 1996.

....of a Semi Autonomous Assistant for Exploratory Data Analysis Robert St. Amant Department of Computer Science North Carolina State University Box 8206 Raleigh, NC 27695 8206 stamant csc.ncsu.edu Paul R. Cohen Computer Science Department, LGRC University of Massachusetts Box 34610 Amherst, MA 01003 4610 cohen cs.umass.edu ABSTRACT Aide is a knowledge based planning assistant for intelligent data exploration that draws on research in mixed initiative planning and collaborative systems. Aide incrementally explores a dataset, guided by user directives and its own evaluation of the data. The ....

....to reflect our understanding of the variables and relationships involved (Cohen 1995) We have eliminated TotalOperations from this model, because, as we know, it is a linear function of the other three variables. The relationship between NavigationOperations and LocalOperations is relatively strong (r = 0:67) as is the relationship between NavigationOperations and ManipulationOperations (r = 0:54) The variables ManipulationOperations and LocalOperations are weakly correlated to begin with (r = 0:29) and if we hold 0.67 0.54 Manipulation Navigation Local Figure 1: Model of operation counts ....

[Article contains additional citation context not shown here]

St. Amant, R., and Cohen, P. R. 1996b. A planner for exploratory data analysis. In Proceedings of the Third International Conference on Artificial Intelligence Planning Systems, 205--212. AAAI Press.

....continues, but this should be enough to give the flavor of the interaction. The analysis and Aide s participation are described in more detail elsewhere [5, 17] MIXED INITIATIVE ASSISTANCE Aide s design exploits a striking similarity between interactive data exploration and planning [18, 19], especially partial hierarchical planning [9] Briefly, a partial hierarchical planner has these properties: A plan library: A great deal of procedural knowledge is not generated from scratch when required, but rather retrieved from memory of past experience. A partial hierarchical planner ....

Robert St. Amant and Paul R. Cohen. A planner for exploratory data analysis. In Proceedings of the Third International Conference on Artificial Intelligence Planning Systems, pages 205--212. AAAI Press, 1996.

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