Action Editor Storyboard

Slides:



Advertisements
Similar presentations
1 KANAL: Knowledge ANALysis Jihie Kim Yolanda Gil USC/ISI
Advertisements

ARCHITECTURES FOR ARTIFICIAL INTELLIGENCE SYSTEMS
Project Proposal.
Conversation Form l One path through a use case that emphasizes interactions between an actor and the system l Can show optional and repeated actions l.
The KB on its way to Web 2.0 Lower the barrier for users to remix the output of services. Theo van Veen, ELAG 2006, April 26.
1 Inheritance. 2 One class inherits from another if it describes a specialized subset of objects Terminology: inheritschild class subclass –the class.
1 Software Testing and Quality Assurance Lecture 15 - Planning for Testing (Chapter 3, A Practical Guide to Testing Object- Oriented Software)
Lesson 6. Refinement of the Operator Model This page describes formally how we refine Figure 2.5 into a more detailed model so that we can connect it.
Task analysis 1 © Copyright De Montfort University 1998 All Rights Reserved Task Analysis Preece et al Chapter 7.
1 USC INFORMATION SCIENCES INSTITUTE Tailor, 6/20/04 TAILOR*: Modifying Calo’s Procedure Knowledge through Instruction Jim Blythe, Yolanda Gil, Jihie Kim.
Semantic Web Technologies Lecture # 2 Faculty of Computer Science, IBA.
Early Design Process Brent M. Dingle, Ph.D Game Design and Development Program Mathematics, Statistics and Computer Science University of Wisconsin.
Authoring and Critiquing COA’s in SHAKEN: A Storyboard showing work in progress UT May 2002.
1 USC INFORMATION SCIENCES INSTITUTE TEMPLE meeting, July 2000 TEMPLE: TEMPLate Enhancement through Knowledge Acquisition Yolanda Gil Jim Blythe Jihie.
Knowledge Entry as the Graphical Assembly of Components Peter Clark, John Thompson (Boeing) Ken Barker, Bruce Porter (Univ Texas at Austin) Vinay Chaudhri,
Rubicon ATLAS A Basic User’s Manual.
Middle States Self-Study Online Resources. Primary Web Resources  Provost’s MSCHE site  Document and Feedback request forms  Secure MSCHE Document.
10/12/ Recall The Team Skills 1. Analyzing the Problem (with 5 steps) 2. Understanding User and Stakeholder Needs 1. Interviews & questionnaires.
1 USC INFORMATION SCIENCES INSTITUTE TEMPLE meeting, July 2000 Specifying Planning Objectives Yolanda Gil Jim Blythe Jihie Kim Surya Ramachandran
COMP106 Assignment 2 Proposal 1. Interface Tasks My new interface design for the University library catalogue will incorporate all of the existing features,
1 USC INFORMATION SCIENCES INSTITUTE CALO, 8/8/03 Acquiring advice (that may use complex expressions) and action specifications Acquiring planning advice,
Introduction CS 3358 Data Structures. What is Computer Science? Computer Science is the study of algorithms, including their  Formal and mathematical.
Using and modifying plan constraints in Constable Jim Blythe and Yolanda Gil Temple project USC Information Sciences Institute
1 USC, INFORMATION SCIENCES INSTITUTE An integrated environment for KA An Integrated Environment for Knowledge Acquisition Jim Blythe
Using and modifying plan constraints in Constable Jim Blythe and Yolanda Gil Temple project USC Information Sciences Institute
OilEd An Introduction to OilEd Sean Bechhofer. Topics we will discuss Basic OilEd use –Defining Classes, Properties and Individuals in an Ontology –This.
1 USC INFORMATION SCIENCES INSTITUTE Expect: COA Critiquing PSM EXPECT: A User-Centered Environment for the Development and Adaptation of Knowledge-Based.
1 USC INFORMATION SCIENCES INSTITUTE EXPECT TEMPLE: TEMPLate Extension Through Knowledge Acquisition Yolanda Gil Jim Blythe Information Sciences Institute.
1 FollowMyLink Individual APT Presentation First Talk February 2006.
Lecture 5 Frames. Associative networks, rules or logic do not provide the ability to group facts into associated clusters or to associate relevant procedural.
KANAL (Knowledge ANALysis) Jihie Kim Jim Blythe Yolanda Gil
Unit 6 : Dimensioning & Parametric Constraints DT2510: Advanced CAD Methods.
1 Introduction to HTML. 2 Definitions  W W W – World Wide Web.  HTML – HyperText Markup Language – The Language of Web Pages on the World Wide Web.
KANAL (Knowledge ANALysis) Status Jihie Kim Yolanda Gil Jim Blythe Varun Ratnakar
IBM Rational Rhapsody Advanced Systems Training v7.5
Strategic Information Systems Planning
Algorithms and Problem Solving
Essentials of UrbanCode Deploy v6.1 QQ147
CMPE 280 Web UI Design and Development August 29 Class Meeting
Dialog Manager for COA entry
KANAL: Knowledge ANALysis
Creating & Editing Reports
Recall The Team Skills Analyzing the Problem (with 5 steps)
Implementing Critiquing Question #3 using Patterns
Building a User Interface with Forms
Use Case Model.
SCC P2P – Collaboration Made Easy Contract Management training
KANAL: Knowledge ANALysis
Material available from the HPKB COA challenge problem
COA critiquing through normative simulation
Object oriented system development life cycle
Abstract descriptions of systems whose requirements are being analysed
Intent (Thanks to Jim Fawcett for the slides)
Understanding Inheritance
Exploring Microsoft® Access® 2016 Series Editor Mary Anne Poatsy
Intro to Expert Systems Paula Matuszek CSC 8750, Fall, 2004
The Smarter Balanced Assessment Consortium
AN INTRODUCTION TO: POWERPOINT.
CHAPTER 4 PROPOSAL.
CHAPTER 4 PROPOSAL.
COA critiquing through normative simulation
Analysis models and design models
2. Ontology development tools/modules
CP Storyboard Proposal
Web Development Using ASP .NET
TEMPLE: TEMPLate Enhancement through Knowledge Acquisition
Applying Use Cases (Chapters 25,26)
Applying Use Cases (Chapters 25,26)
Yolanda Gil Jihie Kim Jim Blythe Surya Ramachandran
Key Updates to the Online IEP
Presentation transcript:

Action Editor Storyboard Please use the “Notes Page View” for details Action Editor Storyboard Jim Blythe Jihie Kim Yolanda Gil USC Information Sciences Institute June 14th, 2002

Adding critiquing knowledge in Shaken The component library is the primary source of background information used in critiquing. Some critiquing knowledge will be added by specializing concepts from the component library or adding new concepts to extend it. Kanal uses knowledge about actions -- preconditions and effects. We can add critiquing knowledge for Kanal by adding new situation-dependent specializations of the actions. This does not modify the component library, but extends it.

Example: a specialized version of ‘destroy’ when the attack is on the flank We assume the action ‘destroy’ has a domain-independent definition in the component library Initially, (through pump-priming) the KB has a special case of ‘destroy’ for military units in which the agent needs a 3:1 force ratio to be successful (e.g. to achieve the expected effect: attrition rate > 40%) However, in a SME-generated COA, a 2.5:1 ratio is sufficient because the attack is in the Red flank. The SME will create a new special case of the ‘destroy’ action so that normative simulation can test mission accomplishment. To check mission accomplishment, a normative simulation would check that all the actions are possible and that they combine to achieve the plan’s desired effects. Here we consider an example where the tool believes the action is not possible because it has an over-constrained model of the preconditions.

Initial critique of mission accomplishment From KANAL: Step: Attack to destroy red forces Checking conditions 1. There is no terrain advantage for red  This condition succeeded 2. Force ratio >= 3:1  This condition failed Click here to find help for fixing this Checking expected effects from this step 1. Red forces are destroyed  the expected effect failed attrition rate < 40% Kanal warns the user that the preconditions are not satisfied for the attack to destroy, and offers several suggestions for fixing this. These will include finding steps to add to the COA to reduce Red’s combat power or increase Blue’s combat power, as well as creating a new specialization of the destroy action. We assume here that the user chooses this last option.

Viewing the subclasses of the ‘destroy’ action The user can see the different subclasses of ‘Destroy’ in Shaken’s hierarchy view. The subclasses represent special cases of the actions, with modified behavior in specific cases. Initially the differences can be implemented using KM properties that influence generally defined effects and preconditions, e.g. ‘required-force-ratio’ We can start from the standard Shaken class hierarchy, where the user can see the subclasses of the Destroy action that are currently defined.

trigger: agent is militaryUnit and object is militaryUnit if: force ratio >= 3:1 and object does not have terrain advantage object attrition is 50% agent attrition is 10% trigger: red has medium terrain advantage object attrition is 35% The ‘triggers’ correspond to the defining elements of the subclass in KM. Here we only show the information that distinguishes an action from its parent, rather than the information distinguishing, for example, Destroy-RedTerrainAdvantage from Destroy. The full defining condition can be found as the conjunction of triggers along the path back to the root. This defining condition can then be used so that KM’s matcher can identify the appropriate action. Triggers can be shown and defined through CMAPS with the general ability to show distinguished nodes and links, as shown in the next slide. We aim to show information to the user about an action that includes the trigger and any important properties that are different from the parent action. The boxes in this slide showing that information could be implemented using the text box that already appears when the mouse moves over an element in the hierarchy, with information about the trigger and properties automatically generated in cases where the user defines the action and there is no associated text description.

The CMAP view shows the trigger as a distinguished set of features attrition required-force-ratio The idea of allowing a distinguished set of nodes and/or links to be viewed and set in the CMAP has been discusses as a generally useful modification. Here it can be used to define the trigger of the action, which is created as a subclass of the parent action. In this slide, the trigger conditions are shown surrounded with a red box, but the actual mechanism doesn’t affect the way it is used here. The two properties shown here are used in the preconditions and effects defined in pump priming on the Destroy-MilitaryUnit action. I’m using green for the property values, following Ken B’s slides. Trigger 3 50%

Define the new special case by its trigger and essential differences from the parent (Destroy-MilitaryUnit) Trigger required-force-ratio point-of-attack flank The user can define a special case of this action by creating a subclass, defining its trigger and modifying the properties as needed. The trigger is used to build the defining elements of the class in KM, so Kanal and other simulators can use the KM matcher to find the appropriate action when Destroy is used in a COA. 2.5

The user can see the new special case in the hierarchy view trigger: agent is militaryUnit and object is militaryUnit if: force ratio >= 3:1 and object does not have terrain advantage object attrition is 50% agent attrition is 10% trigger: red has medium terrain advantage object attrition is 35% The new subclass can be seen in the action hierarchy along with the other subclasses that have been defined. This approach to refining actions for specific situations using subclasses is useful because it avoids encoding the special cases in complex preconditions and conditional effects. Users typically have difficulties with such complex logical expressions. Studies have shown that users tend to describe complex logical formulae in terms of a prototypical case and exceptions (eg Pane et al 01), and the same approach is taken in ripple-down rules (eg Preston et al 94). trigger: point of attack is the flank of the object if: force ratio >= 2.5:1 and object does not have terrain advantage

Revised critique of mission accomplishment KANAL is able to use the appropriate action using KM’s matcher Step: Attack to destroy red forces Checking conditions 1. There is no terrain advantage for red  This condition succeeded 2. Force ratio >= 2.5:1 and the point of attack is the flank of the object Checking expected effects from this step 1. Red forces are destroyed  the expected effect succeeded Kanal notes that the action preconditions are satisfied through a special case, but that the action still succeeds.

Examples requiring this capability Our previous HPKB work on estimating force ratios in COA critiquing demonstrates this requirement. Our previous HPKB work on estimating length of time to repair damage to bridges also demonstrates it. Also previous experience in real-world planning KBs (for oil-spill domain, machining, …) In the above cases, KA was done by KE.

From HPKB experience estimating force ratios Typically require ratio of 3:1 for attack, but only 2.5:1 for attack on units in a ‘hasty defense’. Required force ratio is also reduced if the red forces are making a ‘mobile defense’. Also actual force ratio increased if the red forces are canalized (strung out): can penetrate and only engage 1 sub-unit.

From HPKB experience estimating length of time to repair a bridge This was accomplished through normative simulation of alternative plans to repair damage using different equipment. Main difference: we needed to use an AI planning system. Preconditions on using floating bridges depend on the amount of ‘extra’ gap on river beyond length of bridge Same for fixed (non-floating) bridges, but in a different way