Data Extraction Research Group by waterwolltoremilion

VIEWS: 1 PAGES: 8

									                       TANGO
       Table ANalysis for Generating Ontologies
                                                         fleck                    velter

        repeat:
        1. understand table                                               gonsity          hepth
                                                                           (ld/gg)          (gd)
        2. generate mini-ontology
        3. match with growing ontology
        4. Adjust & merge                                burlam             1.2            120
        until ontology developed
                                                         falder             2.3            230

                                                         multon             2.5            400




                                                                                                   Growing
                                                                                                   Ontology




                                                                    1:*
                                                              has         gonosity
                                                          1
TANGO in a nutshell: TANGO repeatedly turns      fleck                                             velter
raw tables into conceptual mini-ontologies and
integrates them into a growing ontology.                  1
                                                                    1:* hepth
                                                           has
Integrating and Storing Uncertain Data
             Basic Skills:
                                        Nod           Look at     Wave        Point at           Pick up       …
             - movement
                                      Find face        Speak      Hear      Avoid Obstacle        Identify Object    …
             - capability             Localize       Plan path   Navigate   Generate distribution     …
             - logical               Look happy       Imitate    Follow     Maintain Distance        …
Robot:       - behavioral
Azimo                        Greet                                             Safety Layer:           Maintain Distance

                               Find face          Look at                      Default Layer:          Look happy

                                Wave               Nod
                                                                               Task: Play Imitation Game
                                Speak
                                                                                         Greet               Want to play a
                                                                                                           game! Let’s imitate
                             Reward                                                                               her!

                                Identify Apple                                       Speak

Clinician:                      Look at           Point at

Lee                             Speak                                                Imitate                     Reward
Documents   Patterns       Results
            Sorted
                       Aaron David …
               A       Aarons George S …
               B       Abbott Charles H …
               C       …

            Layout
                       W. S. NEWBURY
                       W. H. ADAMS
                       JOSEPH BACHMAN
                       …

            Logical
            -Name      T. M. Gatch
                       E. H. Stolte
             -Title
                       W. S. Newbury
             -City     …
                                                    Human Behavior
                                                  •Fire Escape Placement
                                Government      •Fugitive Chase Simulation
                                Professionals
                                                  •Adaptive Fire-fighting
   Truly
  Dynamic                                            Animal Learning
  Behavior                                    •Parent-Child Learning Transfer
                                   Biologists
  •Graphics                                        •Predator Introduction
  •Machine                                      •Mutual Genetic Adaptation
   Learning
 •Simulation                     Entertainment
                                                   Movies and Games
                                   Industry    •1,000+ Asymmetric Actors
                                                •Human-like AI Adaptation
                                                 •Screen-based Intelligence
Brian Ricks - BYU CS - October, 2009
Incremental Multi-label Classification with Unknown Labels
  Happy      Bright   Peaceful   Dark   Gloomy


                                                                               We don't know which labels we might
                                                     ? ? ?...                  encounter nor how many labels there
                                                                               will be during training.

                                                                               We need to be able to dynamically
                                                                               add new labels into our learning model.




            Feature Extraction


                                 Happy, Bright, Energetic,?,?,?,?, ...
We cannot assume implicit                                                Wet,Happy,?,?,?,...
negativity for images with                                                                     Dark, Gloomy,?,?,?,...
missing labels.
   0/9 algorithms

   Heuristics
        DN=1
        CL=0.02
        DS=0.01
        MV=0

   Type: Outlier

								
To top