SlideShare uma empresa Scribd logo
1 de 40
Baixar para ler offline
TUG-KMI




               Authoring System in TARGET
                   www.reachyourtarget.org


                                 ¨
                           Georg Ottl

                   Knowledge Management Institute
                      Cognitive Science Section


                         April 29, 2010




        ¨
  Georg Ottl               April 29, 2010           Page 1/37
TUG-KMI




Outline


    Research environment

    Competence performance assessment

    Experts competence structure modeler

    Probabilistic graphical models
       Factor graphs




        ¨
  Georg Ottl                    April 29, 2010   Page 2/37
TUG-KMI




Outline


    Research environment

    Competence performance assessment

    Experts competence structure modeler

    Probabilistic graphical models
       Factor graphs




        ¨
  Georg Ottl                    April 29, 2010   Page 3/37
TUG-KMI




Transformative, Adaptive, Responsive and enGaging
EnvironmenT (TARGET)



          Serious game based learning
          environment
          Enterprise Competence
          Development
          Improve competences in the
          project management and
          innovation domain



        ¨
  Georg Ottl                      April 29, 2010    Page 4/37
TUG-KMI




Five key concepts of TARGET




        ¨
  Georg Ottl          April 29, 2010   Page 5/37
TUG-KMI




TARGET Learning Process




        ¨
  Georg Ottl         April 29, 2010   Page 6/37
TUG-KMI




Role of TUG-KMI in TARGET



               TUG-KMI responsible for TARGET learning process
               TUG-KMI responsible for workpackage competence
               development
                   Competence performance assessment component
                   Story adaptation/interventions
               Integration competence development/TARGET learning
               process




        ¨
  Georg Ottl                        April 29, 2010                  Page 7/37
TUG-KMI




Competence performance assessment mockup




        ¨
  Georg Ottl         April 29, 2010        Page 8/37
TUG-KMI




Outline


    Research environment

    Competence performance assessment

    Experts competence structure modeler

    Probabilistic graphical models
       Factor graphs




        ¨
  Georg Ottl                    April 29, 2010   Page 9/37
TUG-KMI




Competence performance assessment




                                    Competence
               Problems             Assessment
                                                 Competences




        ¨
  Georg Ottl              April 29, 2010                       Page 10/37
TUG-KMI




TARGET competence performance assessment
               Interpret observable performance in game experiences in
               regards to a competence state1 .
                   Include motivational state emotional state in interpretation
               Competence assessment as basis for macro and
               microadaptive2 interventions and adaptations.
               Computational model to automatically assess competence
               state.
          1
          Klaus Korossy. “Modeling Knowledge as Competence and Performance”.
    In: Knowledge Spaces: Theories, Empirical Research, Applications. Ed. by
    Dietrich Albert and Josef Lukas. Mahwah, NJ: Lawrence Erlbaum Associates,
    1999, pp. 103–132.
        2
          Dietrich Albert et al. “Microadaptivity within Complex Learning Situations
    - a Personalized Approach based on Competence Structures and Problem
    Spaces”. In: Proceedings of the international Conference on Computers in
    Education (ICCE 2007). 2007.
        ¨
  Georg Ottl                           April 29, 2010                          Page 11/37
TUG-KMI




TARGET competence performance assessment model
authoring



               Interpretation of the game experiences in terms of
               competences can be done by a social community through
               inspection.
                   Creation of a model by using the social community
                   observations input (cold start problem)
               Experts create a model to automatically interpret performance




        ¨
  Georg Ottl                          April 29, 2010                    Page 12/37
TUG-KMI




TARGET competence performance assessment
requirements



               Knowledge model/competence state exchange with HRM
               Systems such as SAP.
               Assessment in realtime3 to enable microadaptive interventions.




       3
         O. Conlan et al. Realtime Knowledge Space Skill Assessment for
    Personalized Digital Educational Games. IEEE, 2009, pp. 538–542.
        ¨
  Georg Ottl                         April 29, 2010                       Page 13/37
TUG-KMI




Basic principle of probabilistic assessment of the
competence state


       1. If the learner has the competence ci , than increase the
          likelihood of all competence states γci containing ci and
          decrease the likelihood of all competence states γ ci .
       2. If the learner does not have the competence ci , than decrease
          the likelihood of all competence states γci containing ci and
          increase the likelihood of all competence states γ ci .




        ¨
  Georg Ottl                     April 29, 2010                       Page 14/37
TUG-KMI




Assessment calculation complexity reduction

               No structure. Possibly 2n states to be updated on every
               performance observation
               Definition of a partial order relation on competences
               exploiting the properties of the “PrerequesiteOf” relation type
               reduces amount of possible competence states to be taken
               into consideration.
               Can experts or the community directly create a competence
               assessment model??
    Authoring tools can help to create a model for competence
    assessment.


        ¨
  Georg Ottl                          April 29, 2010                      Page 15/37
TUG-KMI




Mathematical and computational model for competence
assessment
               Nondeterministic assessment4
               Traditional, multiplicative update rule56
               Belief propagation networks such as Bayesian Networks7
          4
          C. Hockemeyer. “A Comparison of non-deterministic procedures for the
    adaptive assessment of knowledge”. In: Psychologische Beitrage 44.4 (2002),
    pp. 495–503.
        5
          Jean-Claude Falmagne and Jean-Paul Doignon. “A class of stochastic
    procedures for the assessment of knowledge”. In: British Journal of
    Mathematical and Statistical Psychology 41 (1988), pp. 1–23.
        6
          Jean-Claude Falmagne and Jean-Paul Doignon. “A markovian procedure
    for assessing the state of a system”. In: Journal of Mathematical Psychology
    32.3 (1988), pp. 232–258.
        7
          M. Villano. “Probabilistic Student Models: Bayesian Belief Networks and
    Knowledge Space Theory”. In: Proceedings of the Second International
    Conference on Intelligent Tutoring Systems. Springer, 1992, 491–498.
        ¨
  Georg Ottl                         April 29, 2010                           Page 16/37
TUG-KMI




Outline


    Research environment

    Competence performance assessment

    Experts competence structure modeler

    Probabilistic graphical models
       Factor graphs




        ¨
  Georg Ottl                    April 29, 2010   Page 17/37
TUG-KMI




Current state Competence Modeler



               Support to create competence assessment models
               Using the well studied PrerequesiteOf relation8
               Support experts (psychologists) to create knowledge
               structures.




       8
         Dietrich Albert et al. Knowledge Structures. Ed. by Dietrich Albert. New
    York: Springer Verlag, 1994.
        ¨
  Georg Ottl                          April 29, 2010                          Page 18/37
TUG-KMI




Hasse diagram visualization




               Figure: Popular knowledge space visualizations

        ¨
  Georg Ottl                   April 29, 2010                   Page 19/37
TUG-KMI




Current state Competence Modeler
Problems solved and Related Problems




               Computer supported Hasse diagramm creation
               Visualizations done with the Java Universal Network/Graph
               Framework (JUNG) framework9




      9
        J. Madadhain et al. “Analysis and visualization of network data using
    JUNG”. In: Journal of Statistical Software 10 (2005), pp. 1–35.
        ¨
  Georg Ottl                         April 29, 2010                             Page 20/37
TUG-KMI




Hasse diagram visualization




                                →
                  Figure: Version 0.13 and 0.16




        ¨
  Georg Ottl              April 29, 2010          Page 21/37
TUG-KMI




        ¨
  Georg Ottl   April 29, 2010   Page 22/37
TUG-KMI




Current state Competence Modeler
Problems solved and Related Problems

               Creation of a Hasse diagram reduced to the problem of
               calculating the minimal transitive reduction of a graph which
               was shown to have the same complexity as calculation of the
               transitive closure of a graph10 .
               Effective calculation and detection of cycles by maintaining
               the online topological order of the graph11
               Visualizations done with the JUNG12
          10
          A. V. Aho, M. R. Garey, and J. D. Ullman. “The Transitive Reduction of a
    Directed Graph”. In: SIAM Journal on Computing 1.2 (1972), pp. 131–137.
       11
          David J. Pearce and Paul H. J. Kelly. “A dynamic topological sort
    algorithm for directed acyclic graphs”. In: J. Exp. Algorithmics 11 (2006),
    p. 1.7.
       12
          J. Madadhain et al. “Analysis and visualization of network data using
    JUNG”. In: Journal of Statistical Software 10 (2005), pp. 1–35.
        ¨
  Georg Ottl                          April 29, 2010                         Page 23/37
TUG-KMI




Current State Competence Modeler
On-Line Demo Afternoon




               https://dev-css.tu-graz.ac.at/




        ¨
  Georg Ottl                     April 29, 2010   Page 24/37
TUG-KMI




Outline


    Research environment

    Competence performance assessment

    Experts competence structure modeler

    Probabilistic graphical models
       Factor graphs




        ¨
  Georg Ottl                    April 29, 2010   Page 25/37
TUG-KMI




Probabilistic Graphical Models
Whatfor?




               Simple way to visualize the structure of a probabilistic model
               Graphical representation allows insights into the properties of
               the model
                   Insights into conditional independence properties
               Complex computations can be expressed in terms of graphical
               representations; use of graph based inference algorithms that
               exploit graph properties for calculation.




        ¨
  Georg Ottl                           April 29, 2010                      Page 26/37
TUG-KMI




Graph Terminology




          A graph comprises vertices
          V = (a, b, c, d) connected
          by edges




    Definition
    A graph G is a pair G = (V , E ), where V is a (finite) set of
    vertices and E ⊆ V × V is a (finite) set of edges.

        ¨
  Georg Ottl                      April 29, 2010                    Page 27/37
TUG-KMI




    Definition
    A graph G is called undirected iff

                  ∀A, B ∈ V : (A, B) ∈ E ⇒ (B, A) ∈ E             (1)

    Two ordered pairs (A, B) and (B, A) are identified and represented
    by only one undirected edge.

    Definition
    A graph G is called directed iff

                  ∀A, B ∈ V : (A, B) ∈ E ⇒ (B, A) ∈ E             (2)

    An edge (A, B) considered to be a directed edge from A towards
    B

        ¨
  Georg Ottl                   April 29, 2010                    Page 28/37
TUG-KMI




Visual Representation Graph Models




                                                      Figure: Directed Graph
          Figure: Undirected Graph

        ¨
  Georg Ottl                         April 29, 2010                            Page 29/37
TUG-KMI




Probabilistic Graphical Models (1/2)
Whatfor?



               In a probabilistic graph model every vertice represents a
               random variable
               The edges express probabilistic relationships between the
               variables
               Directed Graphical probabilistic Models
                   Bayesian Networks
               Undirected Graphical Probabilistic Models
                   Markov Random Fields
                   Loose coupling between statistical variables.



        ¨
  Georg Ottl                           April 29, 2010                      Page 30/37
TUG-KMI




Graphical probabilistic models


               Question: Can directed probabilistic models such as as
               Bayesian networks be used for assessment. How does believe
               propagation relate to the classical update rule?
               Question: How is the relation between directed and
               undirected probabilistic graphical models?
               Use of directed and undirected graphical probabilistic models
               to assess the players state
               Efficient sum and dot product calculation.




        ¨
  Georg Ottl                          April 29, 2010                     Page 31/37
TUG-KMI




A flexible probabilistic graphical model, the Factor Graph


               Factor Graphs13 as a single representation for directed and
               undirected graphical probabilistic models
               Factor Graphs were successfully applied for Bayesian Networks
               and Markovian Models
               Multiple applications in artificial intelligence and signal
               processing based on Factor Graphs




       13
          Frank Kschischang et al. “Factor Graphs and the Sum-Product Algorithm”.
    In: IEEE Transactions on Information Theory 47 (2001), pp. 498–519.
        ¨
  Georg Ottl                           April 29, 2010                       Page 32/37
TUG-KMI




Factor graph conversion (1/3)




  Example 1:(A simple probabilistic graph)      S1        S2
  Let f (S1, S2, S3) be a function of three
  variables, and suppose that f can be
  expressed as a product
  f (S1, S2, S3) = p(S1)p(S2)p(S3|S1, S2)            S3




        ¨
  Georg Ottl                   April 29, 2010             Page 33/37
TUG-KMI




Factor graph conversion (2/3)


                                                S1        S2

  Example 1:(A factor graph)
  Let f (S1, S2, S3) be a function of three
  variables, and suppose that f can be               f
  expressed as a product
  f (S1, S2, S3) = p(S1)p(S2)p(S3|S1, S2)

                                                     S3



        ¨
  Georg Ottl                   April 29, 2010             Page 34/37
TUG-KMI




Factor graph conversion (3/3)



               Efficient algorithms available to calculate probabilities
               (Sum-Product algorithm)
                   Makes extensive use of “conditional independent” properties
                   Parallelization possible
               Approximative algorithms
               Efficient marginalization




        ¨
  Georg Ottl                          April 29, 2010                        Page 35/37
TUG-KMI




Thank You!




        ¨
  Georg Ottl   April 29, 2010   Page 36/37
TUG-KMI


      [1]      A. V. Aho, M. R. Garey, and J. D. Ullman. “The Transitive
               Reduction of a Directed Graph”. In: SIAM Journal on
               Computing 1.2 (1972), pp. 131–137.
      [2]      Dietrich Albert et al. Knowledge Structures. Ed. by
               Dietrich Albert. New York: Springer Verlag, 1994.
      [3]      Dietrich Albert et al. “Microadaptivity within Complex
               Learning Situations - a Personalized Approach based on
               Competence Structures and Problem Spaces”. In:
               Proceedings of the international Conference on Computers in
               Education (ICCE 2007). 2007.
      [4]      O. Conlan et al. Realtime Knowledge Space Skill Assessment
               for Personalized Digital Educational Games. IEEE, 2009,
               pp. 538–542.
      [5]      Jean-Claude Falmagne and Jean-Paul Doignon. “A class of
               stochastic procedures for the assessment of knowledge”. In:
        ¨
  Georg Ottl                        April 29, 2010                     Page 36/37
TUG-KMI


               British Journal of Mathematical and Statistical Psychology
               41 (1988), pp. 1–23.
      [6]      Jean-Claude Falmagne and Jean-Paul Doignon. “A
               markovian procedure for assessing the state of a system”. In:
               Journal of Mathematical Psychology 32.3 (1988),
               pp. 232–258.
      [7]      C. Hockemeyer. “A Comparison of non-deterministic
               procedures for the adaptive assessment of knowledge”. In:
               Psychologische Beitrage 44.4 (2002), pp. 495–503.
      [8]      Klaus Korossy. “Modeling Knowledge as Competence and
               Performance”. In: Knowledge Spaces: Theories, Empirical
               Research, Applications. Ed. by Dietrich Albert and
               Josef Lukas. Mahwah, NJ: Lawrence Erlbaum Associates,
               1999, pp. 103–132.


        ¨
  Georg Ottl                        April 29, 2010                      Page 36/37
TUG-KMI


      [9]      Frank Kschischang et al. “Factor Graphs and the
               Sum-Product Algorithm”. In: IEEE Transactions on
               Information Theory 47 (2001), pp. 498–519.

    [10]       J. Madadhain et al. “Analysis and visualization of network
               data using JUNG”. In: Journal of Statistical Software 10
               (2005), pp. 1–35.

    [11]       David J. Pearce and Paul H. J. Kelly. “A dynamic
               topological sort algorithm for directed acyclic graphs”. In: J.
               Exp. Algorithmics 11 (2006), p. 1.7.

    [12]       M. Villano. “Probabilistic Student Models: Bayesian Belief
               Networks and Knowledge Space Theory”. In: Proceedings of
               the Second International Conference on Intelligent Tutoring
               Systems. Springer, 1992, 491–498.
        ¨
  Georg Ottl                         April 29, 2010                       Page 37/37
TUG-KMI




Acronyms




    JUNG Java Universal Network/Graph Framework




        ¨
  Georg Ottl               April 29, 2010         Page 37/37

Mais conteúdo relacionado

Semelhante a Authoring System in TARGET

Document.doc.doc
Document.doc.docDocument.doc.doc
Document.doc.docbutest
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender SystemsMarcel Kurovski
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systemsinovex GmbH
 
Introduction to LLMs
Introduction to LLMsIntroduction to LLMs
Introduction to LLMsLoic Merckel
 
Easing the Reuse of ML Solutions by Interactive Clustering-based Autotuning i...
Easing the Reuse of ML Solutions by Interactive Clustering-based Autotuning i...Easing the Reuse of ML Solutions by Interactive Clustering-based Autotuning i...
Easing the Reuse of ML Solutions by Interactive Clustering-based Autotuning i...SEAA 2022
 
Mental Model for Exploratory Data Analysis Applications for Structured Proble...
Mental Model for Exploratory Data Analysis Applications for Structured Proble...Mental Model for Exploratory Data Analysis Applications for Structured Proble...
Mental Model for Exploratory Data Analysis Applications for Structured Proble...Jukka-Matti Turtiainen
 
AsianPLoP'14: How and Why Design Patterns Impact Quality and Future Challenges
AsianPLoP'14: How and Why Design Patterns Impact Quality and Future ChallengesAsianPLoP'14: How and Why Design Patterns Impact Quality and Future Challenges
AsianPLoP'14: How and Why Design Patterns Impact Quality and Future ChallengesPtidej Team
 
GDG Cloud Southlake #17: Meg Dickey-Kurdziolek: Explainable AI is for Everyone
GDG Cloud Southlake #17: Meg Dickey-Kurdziolek: Explainable AI is for EveryoneGDG Cloud Southlake #17: Meg Dickey-Kurdziolek: Explainable AI is for Everyone
GDG Cloud Southlake #17: Meg Dickey-Kurdziolek: Explainable AI is for EveryoneJames Anderson
 
ARgh! kinesthetic learning
ARgh! kinesthetic learningARgh! kinesthetic learning
ARgh! kinesthetic learningfridolin.wild
 
An Embeddable Dashboard for Widget-Based Visual Analytics on Scientific Commu...
An Embeddable Dashboard for Widget-Based Visual Analytics on Scientific Commu...An Embeddable Dashboard for Widget-Based Visual Analytics on Scientific Commu...
An Embeddable Dashboard for Widget-Based Visual Analytics on Scientific Commu...Michael Derntl
 
Applied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDLApplied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDLMarc Teunis
 
Daniel Samaan: ChatGPT and the Future of Work
Daniel Samaan: ChatGPT and the Future of WorkDaniel Samaan: ChatGPT and the Future of Work
Daniel Samaan: ChatGPT and the Future of WorkEdunomica
 
Big-Data Analytics for Media Management
Big-Data Analytics for Media ManagementBig-Data Analytics for Media Management
Big-Data Analytics for Media Managementtechkrish
 
Panel at acm_sigplan_trust2014
Panel at acm_sigplan_trust2014Panel at acm_sigplan_trust2014
Panel at acm_sigplan_trust2014Grigori Fursin
 
ARTIFICIAL INTELLIGENCE ppt.pptx
ARTIFICIAL INTELLIGENCE ppt.pptxARTIFICIAL INTELLIGENCE ppt.pptx
ARTIFICIAL INTELLIGENCE ppt.pptxVinodhKumar658343
 
Partial Object Detection in Inclined Weather Conditions
Partial Object Detection in Inclined Weather ConditionsPartial Object Detection in Inclined Weather Conditions
Partial Object Detection in Inclined Weather ConditionsIRJET Journal
 
Capacity-Driven Scaling Schedules Derivation or Coordinated Elasticity of Con...
Capacity-Driven Scaling Schedules Derivation or Coordinated Elasticity of Con...Capacity-Driven Scaling Schedules Derivation or Coordinated Elasticity of Con...
Capacity-Driven Scaling Schedules Derivation or Coordinated Elasticity of Con...Vladimir Podolskiy
 
DynaLearn@JTEL2010_2010_6_9
DynaLearn@JTEL2010_2010_6_9DynaLearn@JTEL2010_2010_6_9
DynaLearn@JTEL2010_2010_6_9Wouter Beek
 

Semelhante a Authoring System in TARGET (20)

Document.doc.doc
Document.doc.docDocument.doc.doc
Document.doc.doc
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Introduction to LLMs
Introduction to LLMsIntroduction to LLMs
Introduction to LLMs
 
Easing the Reuse of ML Solutions by Interactive Clustering-based Autotuning i...
Easing the Reuse of ML Solutions by Interactive Clustering-based Autotuning i...Easing the Reuse of ML Solutions by Interactive Clustering-based Autotuning i...
Easing the Reuse of ML Solutions by Interactive Clustering-based Autotuning i...
 
Mental Model for Exploratory Data Analysis Applications for Structured Proble...
Mental Model for Exploratory Data Analysis Applications for Structured Proble...Mental Model for Exploratory Data Analysis Applications for Structured Proble...
Mental Model for Exploratory Data Analysis Applications for Structured Proble...
 
AsianPLoP'14: How and Why Design Patterns Impact Quality and Future Challenges
AsianPLoP'14: How and Why Design Patterns Impact Quality and Future ChallengesAsianPLoP'14: How and Why Design Patterns Impact Quality and Future Challenges
AsianPLoP'14: How and Why Design Patterns Impact Quality and Future Challenges
 
GDG Cloud Southlake #17: Meg Dickey-Kurdziolek: Explainable AI is for Everyone
GDG Cloud Southlake #17: Meg Dickey-Kurdziolek: Explainable AI is for EveryoneGDG Cloud Southlake #17: Meg Dickey-Kurdziolek: Explainable AI is for Everyone
GDG Cloud Southlake #17: Meg Dickey-Kurdziolek: Explainable AI is for Everyone
 
ARgh! kinesthetic learning
ARgh! kinesthetic learningARgh! kinesthetic learning
ARgh! kinesthetic learning
 
An Embeddable Dashboard for Widget-Based Visual Analytics on Scientific Commu...
An Embeddable Dashboard for Widget-Based Visual Analytics on Scientific Commu...An Embeddable Dashboard for Widget-Based Visual Analytics on Scientific Commu...
An Embeddable Dashboard for Widget-Based Visual Analytics on Scientific Commu...
 
Applied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDLApplied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDL
 
202212APSEC.pptx.pdf
202212APSEC.pptx.pdf202212APSEC.pptx.pdf
202212APSEC.pptx.pdf
 
GazeObjectDetection.pptx
GazeObjectDetection.pptxGazeObjectDetection.pptx
GazeObjectDetection.pptx
 
Daniel Samaan: ChatGPT and the Future of Work
Daniel Samaan: ChatGPT and the Future of WorkDaniel Samaan: ChatGPT and the Future of Work
Daniel Samaan: ChatGPT and the Future of Work
 
Big-Data Analytics for Media Management
Big-Data Analytics for Media ManagementBig-Data Analytics for Media Management
Big-Data Analytics for Media Management
 
Panel at acm_sigplan_trust2014
Panel at acm_sigplan_trust2014Panel at acm_sigplan_trust2014
Panel at acm_sigplan_trust2014
 
ARTIFICIAL INTELLIGENCE ppt.pptx
ARTIFICIAL INTELLIGENCE ppt.pptxARTIFICIAL INTELLIGENCE ppt.pptx
ARTIFICIAL INTELLIGENCE ppt.pptx
 
Partial Object Detection in Inclined Weather Conditions
Partial Object Detection in Inclined Weather ConditionsPartial Object Detection in Inclined Weather Conditions
Partial Object Detection in Inclined Weather Conditions
 
Capacity-Driven Scaling Schedules Derivation or Coordinated Elasticity of Con...
Capacity-Driven Scaling Schedules Derivation or Coordinated Elasticity of Con...Capacity-Driven Scaling Schedules Derivation or Coordinated Elasticity of Con...
Capacity-Driven Scaling Schedules Derivation or Coordinated Elasticity of Con...
 
DynaLearn@JTEL2010_2010_6_9
DynaLearn@JTEL2010_2010_6_9DynaLearn@JTEL2010_2010_6_9
DynaLearn@JTEL2010_2010_6_9
 

Authoring System in TARGET

  • 1. TUG-KMI Authoring System in TARGET www.reachyourtarget.org ¨ Georg Ottl Knowledge Management Institute Cognitive Science Section April 29, 2010 ¨ Georg Ottl April 29, 2010 Page 1/37
  • 2. TUG-KMI Outline Research environment Competence performance assessment Experts competence structure modeler Probabilistic graphical models Factor graphs ¨ Georg Ottl April 29, 2010 Page 2/37
  • 3. TUG-KMI Outline Research environment Competence performance assessment Experts competence structure modeler Probabilistic graphical models Factor graphs ¨ Georg Ottl April 29, 2010 Page 3/37
  • 4. TUG-KMI Transformative, Adaptive, Responsive and enGaging EnvironmenT (TARGET) Serious game based learning environment Enterprise Competence Development Improve competences in the project management and innovation domain ¨ Georg Ottl April 29, 2010 Page 4/37
  • 5. TUG-KMI Five key concepts of TARGET ¨ Georg Ottl April 29, 2010 Page 5/37
  • 6. TUG-KMI TARGET Learning Process ¨ Georg Ottl April 29, 2010 Page 6/37
  • 7. TUG-KMI Role of TUG-KMI in TARGET TUG-KMI responsible for TARGET learning process TUG-KMI responsible for workpackage competence development Competence performance assessment component Story adaptation/interventions Integration competence development/TARGET learning process ¨ Georg Ottl April 29, 2010 Page 7/37
  • 8. TUG-KMI Competence performance assessment mockup ¨ Georg Ottl April 29, 2010 Page 8/37
  • 9. TUG-KMI Outline Research environment Competence performance assessment Experts competence structure modeler Probabilistic graphical models Factor graphs ¨ Georg Ottl April 29, 2010 Page 9/37
  • 10. TUG-KMI Competence performance assessment Competence Problems Assessment Competences ¨ Georg Ottl April 29, 2010 Page 10/37
  • 11. TUG-KMI TARGET competence performance assessment Interpret observable performance in game experiences in regards to a competence state1 . Include motivational state emotional state in interpretation Competence assessment as basis for macro and microadaptive2 interventions and adaptations. Computational model to automatically assess competence state. 1 Klaus Korossy. “Modeling Knowledge as Competence and Performance”. In: Knowledge Spaces: Theories, Empirical Research, Applications. Ed. by Dietrich Albert and Josef Lukas. Mahwah, NJ: Lawrence Erlbaum Associates, 1999, pp. 103–132. 2 Dietrich Albert et al. “Microadaptivity within Complex Learning Situations - a Personalized Approach based on Competence Structures and Problem Spaces”. In: Proceedings of the international Conference on Computers in Education (ICCE 2007). 2007. ¨ Georg Ottl April 29, 2010 Page 11/37
  • 12. TUG-KMI TARGET competence performance assessment model authoring Interpretation of the game experiences in terms of competences can be done by a social community through inspection. Creation of a model by using the social community observations input (cold start problem) Experts create a model to automatically interpret performance ¨ Georg Ottl April 29, 2010 Page 12/37
  • 13. TUG-KMI TARGET competence performance assessment requirements Knowledge model/competence state exchange with HRM Systems such as SAP. Assessment in realtime3 to enable microadaptive interventions. 3 O. Conlan et al. Realtime Knowledge Space Skill Assessment for Personalized Digital Educational Games. IEEE, 2009, pp. 538–542. ¨ Georg Ottl April 29, 2010 Page 13/37
  • 14. TUG-KMI Basic principle of probabilistic assessment of the competence state 1. If the learner has the competence ci , than increase the likelihood of all competence states γci containing ci and decrease the likelihood of all competence states γ ci . 2. If the learner does not have the competence ci , than decrease the likelihood of all competence states γci containing ci and increase the likelihood of all competence states γ ci . ¨ Georg Ottl April 29, 2010 Page 14/37
  • 15. TUG-KMI Assessment calculation complexity reduction No structure. Possibly 2n states to be updated on every performance observation Definition of a partial order relation on competences exploiting the properties of the “PrerequesiteOf” relation type reduces amount of possible competence states to be taken into consideration. Can experts or the community directly create a competence assessment model?? Authoring tools can help to create a model for competence assessment. ¨ Georg Ottl April 29, 2010 Page 15/37
  • 16. TUG-KMI Mathematical and computational model for competence assessment Nondeterministic assessment4 Traditional, multiplicative update rule56 Belief propagation networks such as Bayesian Networks7 4 C. Hockemeyer. “A Comparison of non-deterministic procedures for the adaptive assessment of knowledge”. In: Psychologische Beitrage 44.4 (2002), pp. 495–503. 5 Jean-Claude Falmagne and Jean-Paul Doignon. “A class of stochastic procedures for the assessment of knowledge”. In: British Journal of Mathematical and Statistical Psychology 41 (1988), pp. 1–23. 6 Jean-Claude Falmagne and Jean-Paul Doignon. “A markovian procedure for assessing the state of a system”. In: Journal of Mathematical Psychology 32.3 (1988), pp. 232–258. 7 M. Villano. “Probabilistic Student Models: Bayesian Belief Networks and Knowledge Space Theory”. In: Proceedings of the Second International Conference on Intelligent Tutoring Systems. Springer, 1992, 491–498. ¨ Georg Ottl April 29, 2010 Page 16/37
  • 17. TUG-KMI Outline Research environment Competence performance assessment Experts competence structure modeler Probabilistic graphical models Factor graphs ¨ Georg Ottl April 29, 2010 Page 17/37
  • 18. TUG-KMI Current state Competence Modeler Support to create competence assessment models Using the well studied PrerequesiteOf relation8 Support experts (psychologists) to create knowledge structures. 8 Dietrich Albert et al. Knowledge Structures. Ed. by Dietrich Albert. New York: Springer Verlag, 1994. ¨ Georg Ottl April 29, 2010 Page 18/37
  • 19. TUG-KMI Hasse diagram visualization Figure: Popular knowledge space visualizations ¨ Georg Ottl April 29, 2010 Page 19/37
  • 20. TUG-KMI Current state Competence Modeler Problems solved and Related Problems Computer supported Hasse diagramm creation Visualizations done with the Java Universal Network/Graph Framework (JUNG) framework9 9 J. Madadhain et al. “Analysis and visualization of network data using JUNG”. In: Journal of Statistical Software 10 (2005), pp. 1–35. ¨ Georg Ottl April 29, 2010 Page 20/37
  • 21. TUG-KMI Hasse diagram visualization → Figure: Version 0.13 and 0.16 ¨ Georg Ottl April 29, 2010 Page 21/37
  • 22. TUG-KMI ¨ Georg Ottl April 29, 2010 Page 22/37
  • 23. TUG-KMI Current state Competence Modeler Problems solved and Related Problems Creation of a Hasse diagram reduced to the problem of calculating the minimal transitive reduction of a graph which was shown to have the same complexity as calculation of the transitive closure of a graph10 . Effective calculation and detection of cycles by maintaining the online topological order of the graph11 Visualizations done with the JUNG12 10 A. V. Aho, M. R. Garey, and J. D. Ullman. “The Transitive Reduction of a Directed Graph”. In: SIAM Journal on Computing 1.2 (1972), pp. 131–137. 11 David J. Pearce and Paul H. J. Kelly. “A dynamic topological sort algorithm for directed acyclic graphs”. In: J. Exp. Algorithmics 11 (2006), p. 1.7. 12 J. Madadhain et al. “Analysis and visualization of network data using JUNG”. In: Journal of Statistical Software 10 (2005), pp. 1–35. ¨ Georg Ottl April 29, 2010 Page 23/37
  • 24. TUG-KMI Current State Competence Modeler On-Line Demo Afternoon https://dev-css.tu-graz.ac.at/ ¨ Georg Ottl April 29, 2010 Page 24/37
  • 25. TUG-KMI Outline Research environment Competence performance assessment Experts competence structure modeler Probabilistic graphical models Factor graphs ¨ Georg Ottl April 29, 2010 Page 25/37
  • 26. TUG-KMI Probabilistic Graphical Models Whatfor? Simple way to visualize the structure of a probabilistic model Graphical representation allows insights into the properties of the model Insights into conditional independence properties Complex computations can be expressed in terms of graphical representations; use of graph based inference algorithms that exploit graph properties for calculation. ¨ Georg Ottl April 29, 2010 Page 26/37
  • 27. TUG-KMI Graph Terminology A graph comprises vertices V = (a, b, c, d) connected by edges Definition A graph G is a pair G = (V , E ), where V is a (finite) set of vertices and E ⊆ V × V is a (finite) set of edges. ¨ Georg Ottl April 29, 2010 Page 27/37
  • 28. TUG-KMI Definition A graph G is called undirected iff ∀A, B ∈ V : (A, B) ∈ E ⇒ (B, A) ∈ E (1) Two ordered pairs (A, B) and (B, A) are identified and represented by only one undirected edge. Definition A graph G is called directed iff ∀A, B ∈ V : (A, B) ∈ E ⇒ (B, A) ∈ E (2) An edge (A, B) considered to be a directed edge from A towards B ¨ Georg Ottl April 29, 2010 Page 28/37
  • 29. TUG-KMI Visual Representation Graph Models Figure: Directed Graph Figure: Undirected Graph ¨ Georg Ottl April 29, 2010 Page 29/37
  • 30. TUG-KMI Probabilistic Graphical Models (1/2) Whatfor? In a probabilistic graph model every vertice represents a random variable The edges express probabilistic relationships between the variables Directed Graphical probabilistic Models Bayesian Networks Undirected Graphical Probabilistic Models Markov Random Fields Loose coupling between statistical variables. ¨ Georg Ottl April 29, 2010 Page 30/37
  • 31. TUG-KMI Graphical probabilistic models Question: Can directed probabilistic models such as as Bayesian networks be used for assessment. How does believe propagation relate to the classical update rule? Question: How is the relation between directed and undirected probabilistic graphical models? Use of directed and undirected graphical probabilistic models to assess the players state Efficient sum and dot product calculation. ¨ Georg Ottl April 29, 2010 Page 31/37
  • 32. TUG-KMI A flexible probabilistic graphical model, the Factor Graph Factor Graphs13 as a single representation for directed and undirected graphical probabilistic models Factor Graphs were successfully applied for Bayesian Networks and Markovian Models Multiple applications in artificial intelligence and signal processing based on Factor Graphs 13 Frank Kschischang et al. “Factor Graphs and the Sum-Product Algorithm”. In: IEEE Transactions on Information Theory 47 (2001), pp. 498–519. ¨ Georg Ottl April 29, 2010 Page 32/37
  • 33. TUG-KMI Factor graph conversion (1/3) Example 1:(A simple probabilistic graph) S1 S2 Let f (S1, S2, S3) be a function of three variables, and suppose that f can be expressed as a product f (S1, S2, S3) = p(S1)p(S2)p(S3|S1, S2) S3 ¨ Georg Ottl April 29, 2010 Page 33/37
  • 34. TUG-KMI Factor graph conversion (2/3) S1 S2 Example 1:(A factor graph) Let f (S1, S2, S3) be a function of three variables, and suppose that f can be f expressed as a product f (S1, S2, S3) = p(S1)p(S2)p(S3|S1, S2) S3 ¨ Georg Ottl April 29, 2010 Page 34/37
  • 35. TUG-KMI Factor graph conversion (3/3) Efficient algorithms available to calculate probabilities (Sum-Product algorithm) Makes extensive use of “conditional independent” properties Parallelization possible Approximative algorithms Efficient marginalization ¨ Georg Ottl April 29, 2010 Page 35/37
  • 36. TUG-KMI Thank You! ¨ Georg Ottl April 29, 2010 Page 36/37
  • 37. TUG-KMI [1] A. V. Aho, M. R. Garey, and J. D. Ullman. “The Transitive Reduction of a Directed Graph”. In: SIAM Journal on Computing 1.2 (1972), pp. 131–137. [2] Dietrich Albert et al. Knowledge Structures. Ed. by Dietrich Albert. New York: Springer Verlag, 1994. [3] Dietrich Albert et al. “Microadaptivity within Complex Learning Situations - a Personalized Approach based on Competence Structures and Problem Spaces”. In: Proceedings of the international Conference on Computers in Education (ICCE 2007). 2007. [4] O. Conlan et al. Realtime Knowledge Space Skill Assessment for Personalized Digital Educational Games. IEEE, 2009, pp. 538–542. [5] Jean-Claude Falmagne and Jean-Paul Doignon. “A class of stochastic procedures for the assessment of knowledge”. In: ¨ Georg Ottl April 29, 2010 Page 36/37
  • 38. TUG-KMI British Journal of Mathematical and Statistical Psychology 41 (1988), pp. 1–23. [6] Jean-Claude Falmagne and Jean-Paul Doignon. “A markovian procedure for assessing the state of a system”. In: Journal of Mathematical Psychology 32.3 (1988), pp. 232–258. [7] C. Hockemeyer. “A Comparison of non-deterministic procedures for the adaptive assessment of knowledge”. In: Psychologische Beitrage 44.4 (2002), pp. 495–503. [8] Klaus Korossy. “Modeling Knowledge as Competence and Performance”. In: Knowledge Spaces: Theories, Empirical Research, Applications. Ed. by Dietrich Albert and Josef Lukas. Mahwah, NJ: Lawrence Erlbaum Associates, 1999, pp. 103–132. ¨ Georg Ottl April 29, 2010 Page 36/37
  • 39. TUG-KMI [9] Frank Kschischang et al. “Factor Graphs and the Sum-Product Algorithm”. In: IEEE Transactions on Information Theory 47 (2001), pp. 498–519. [10] J. Madadhain et al. “Analysis and visualization of network data using JUNG”. In: Journal of Statistical Software 10 (2005), pp. 1–35. [11] David J. Pearce and Paul H. J. Kelly. “A dynamic topological sort algorithm for directed acyclic graphs”. In: J. Exp. Algorithmics 11 (2006), p. 1.7. [12] M. Villano. “Probabilistic Student Models: Bayesian Belief Networks and Knowledge Space Theory”. In: Proceedings of the Second International Conference on Intelligent Tutoring Systems. Springer, 1992, 491–498. ¨ Georg Ottl April 29, 2010 Page 37/37
  • 40. TUG-KMI Acronyms JUNG Java Universal Network/Graph Framework ¨ Georg Ottl April 29, 2010 Page 37/37