168 research outputs found
Scalable and Adaptive Goal Recognition
No full textScalable and Adaptive Goal Recognition by Neal Lesh Chairperson of Supervisory Committee: Professor Oren Etzioni Department of Computer Science and Engineering Goal recognition is the task of inferring people's goals based on observing them try to achieve their goals. In software domains, goal recognition can be used to enhance automatic help systems and enable software agents to assist people with their current tasks. Traditionally, research in plan and goal recognition has investigated examples containing less than one hundred plans and goals. We present three novel methods, based on work in machine learning and planning, for scaling up goal recognition to handle on the order of 100,000 candidate goals. First, we show how to automatically construct the plan library from domain primitives, a task previously performed by hand by a human expert. Second, we present compact representations and efficient algorithms for goal recognition. Third, we describe an unsupervised learning algorith..
Interactive Partitioning System Demonstration, Short
No full textPartitioning is often used to support better graph drawing; in this paper, we describe an interactive system in which graph drawing is used to support better partitioning. In our system the user is presented with a drawing of a current network partitioning, and is responsible for choosing appropriate optimization procedures and for focusing their application on portions of the network. Our pilot experiments show that our network drawings succeed in conveying some of the information needed by the human operator to steer the computation effectively, and suggest that interactive, human-guided search may be a useful alternative to fully automatic methods for network and graph partitioning
Fast, Adaptive, and Empirically-Tested Goal Recognition.
No full textCurrent plan recognition research cannot be directly incorporated into real world applications because plan recognizers typically run in exponential time, require a complete plan library as input, and make strong assumptions about the actor's behavior. The objective of the research outlined in this paper is to produce a fast and adaptive goal recognizer that performs well in real domains. Additionally, the goal recognizer must learn its own model of the actor's behavior, instead of receiving a complete plan library. Our recognizer is designed to work on "noisy" streams of executed actions, which represent the ongoing behavior of a suboptimal, forgetful human actor who works on multiple goals at once, abandons plans, executes plans that might not achieve their goal, executes spurious actions, and learns new actions and plans. Introduction Plan recognition (e.g. (Carberry 1990; Pollack 1990)), the task of identifying actors' plans and goals given a partial view of their behavior, is ess..
Interactive data summarization: An example application
No full textSummarizing large multidimensional datasets is a challenging task, often requiring extensive investigation by a user to identify overall trends and important exceptions to them. While many visualization tools help a user produce a single summary of the data at a time, they require the user to explore the dataset manually. Our idea is to have the computer perform an exhaustive search and inform the user about where further investigation is warranted. Our algorithm takes a large, multidimensional dataset as input, along with a specification of the user’s goals, and produces a concise summary that can be clearly visualized in bar graphs or linegraphs. We demonstrate our techniques in a sample prototype for summarizing information stored in spreadsheet databases
OpenROSA Meeting 2009 : report on the 4th OpenROSA, July 20-23, Dar es Salaam, Tanzania; final technical report
No full textThe meeting emphasized networking opportunities and training for people new to the OpenROSA consortium and its projects. OpenROSA (www.openrosa.org) is a group of organizations working together to foster open source, standards-based tools for mobile data collection, aggregation, analysis, and reporting, with a strong emphasis on health applications in low- and middle-income countries. One of its primary outputs has been the JavaROSA codebase (available at code.javarosa.org) for data collection on a wide range of Java-enabled phones. OpenROSA and JavaROSA were launched with funding from the Canadian International Development Research Centre (IDRC)
OpenROSA meeting 2009 : report on the 4th OpenROSA, July 20-23, Dar es Salaam, Tanzania
No full textMeeting: 4th OpenROSA Meeting, 20-23 July, 2009, Dar es Salaam, TZThis is a report on the 4-day conference that brought together international experts. The OpenROSA consortium is a group of organizations working together to foster open source, standards-based tools for mobile data collection, aggregation, analysis, and reporting. JavaRosa is a Java library for rendering forms that are compliant with Open Data Kits (ODK XForms spec). It is the heart of many ODK tools. The emergence of JavaROSA has greatly reduced the duplication of efforts among groups developing solutions for mobile phone-based data collection. OpenROSA and JavaROSA were launched with funding from the Canadian International Development Research Centre (IDRC)
Using Electronic Technology to Improve Clinical Care -- Results from a Before-after Cluster Trial to Evaluate Assessment and Classification of Sick Children According to Integrated Management of Childhood Illness (IMCI) Protocol in Tanzania.
No full textPoor adherence to the Integrated Management of Childhood Illness (IMCI) protocol reduces the potential impact on under-five morbidity and mortality. Electronic technology could improve adherence; however there are few studies demonstrating the benefits of such technology in a resource-poor settings. This study estimates the impact of electronic technology on adherence to the IMCI protocols as compared to the current paper-based protocols in Tanzania. In four districts in Tanzania, 18 clinics were randomly selected for inclusion. At each site, observers documented critical parts of the clinical assessment of children aged 2 months to 5 years. The first set of observations occurred during examination of children using paper-based IMCI (pIMCI) and the next set of observations occurred during examination using the electronic IMCI (eIMCI). Children were re-examined by an IMCI expert and the diagnoses were compared. A total of 1221 children (671 paper, 550 electronic) were observed. For all ten critical IMCI items included in both systems, adherence to the protocol was greater for eIMCI than for pIMCI. The proportion assessed under pIMCI ranged from 61% to 98% compared to 92% to 100% under eIMCI (p < 0.05 for each of the ten assessment items). Use of electronic systems improved the completeness of assessment of children with acute illness in Tanzania. With the before-after nature of the design, potential for temporal confounding is the primary limitation. However, the data collection for both phases occurred over a short period (one month) and so temporal confounding was expected to be minimal. The results suggest that the use of electronic IMCI protocols can improve the completeness and consistency of clinical assessments and future studies will examine the long-term health and health systems impact of eIMCI
Teaching History Now: The University Context: Reflections from the Field
No full textA special section with the following contributors:
Todd Beach, Eastview High School, Apple Valley, MNKristy Brugar, University of OklahomaLendol Calder, Augustana CollegeKaren Carroll Cave, National Humanities CenterFrederick D. Drake, Illinois State UniversityStephen Kneeshaw, College of the OzarksBruce Lesh, Maryland State Department of EducationJodie Mader, Thomas More CollegeDonn Neal, National Archives and Records AdministrationPamela Riney-Kehrberg, Iowa State UniversityRaymond Screws, Arkansas National Guard MuseumWilson J. Warren, Western Michigan Universit
Simulation-based inference for plan monitoring
No full textThe dynamic execution of plans in uncertain domains requires the ability to infer likely current and future world states from past observations. This task can be cast as inference on Dynamic Belief Networks (DBNs) but the resulting networks are difficult to solve with exact methods. We investigate and extend simulation algorithms for approximate inference on Bayesian networks and a propose a new algorithm, called Rewind/Replay, for generating a set of simulations weighted by their likelihood given past observations. We validate our algorithm on a DBN containing thousands of variables, which models the spread of wildfire
- …
