Online Educa 2005 Workshop

boorishadamantΤεχνίτη Νοημοσύνη και Ρομποτική

29 Οκτ 2013 (πριν από 3 χρόνια και 5 μήνες)

53 εμφανίσεις

Dr. Thomas Richter

Online Educa 2005

Workshop

Intelligent Agents for
User Guidance in
Virtual Labs

University of Technology Berlin

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Prospects for Intelligent Assistant
Technologies in eLearning environments for
Mathematics:



Mathematics as science of structure



precisely formalized language



strictly defined entities



well
-
defined dependencies



high continuity of knowlegde

Benefit: Highly structured field

1.
Applied sciences require
mathematics as key technology

2.
Mathematics as a research
-
field
of its own



high level of adaptivity
required

Challenge: Broad audience

algorithmically exploitable

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Mumie


Fields of Learning:



Courses from granular elements of knowledge



Composition with the CourseCreator tool



Interactive multimedia elements



Nonlinear navigation



Exercises, combined into exercise paths



Interactive, constructive



Embedded in an exercise network



Intelligent input & control mechanisms



User driven information retrieval system



Knowledge networks



User defined constructions



Includes an „encyclopaedia“

VirtLabs



Combinable experiments



Explorative learning and research



Experiments integrating CAS & Num. Tools



Intelligent input & control mechanisms

Retrieval

Training

Content

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Layered Course Management: Three Tiered Design

Course in the Content Area

Exercises in the Training Area

Matrix

Determinants

Compute

det(...)

Laplace

Gauss

Algorithm

requires

requires

recommends

Relations between
Exercises

Dependencies of Knowledge
Atoms

requires…

…or requires

Kernel

Gauss

Algorithm

requires

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Training Area


Exercise Network:

Solve y‘ = Ay for ...

Compute exp(A) for ...

Diogonalize A for ...

Alternative Paths

Several prequisites

Exponential Ansatz

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Layered Course Management: Three Tiered Design

Course in the Content Area

Exercises in the Training Area

Asset Level (e.g Virtual Lab Area)

Matrix

Determinants

Compute

det(...)

Laplace

Gauss

Algorithm

Solve

equations

Build

Triangular

Matrix

What is

triangular?

What is a


matrix?

Elementary

Operations

requires

requires

recommends

Relations between
Exercises

Dependencies of Knowledge
Atoms

Storyboarding of
Training Units

requires…

…or requires

Kernel

Gauss

Algorithm

requires

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

What is a „Virtual Laboratory“ ?

Attempt of a Definition:



Virtual Labs are software environments that use
the metaphor of real labs; they allow to design,
setup and carry out experiments by means of
the computer.


Experiments are typically run on computer
implemented abstract algorithms rather than
real objects; these algorithms model either real
devices and items, or theoretical concepts and
objects.

Virtuality


(runs in a computer)

Complexity



(broadness of the field)

Flexibility

Allows composition of
components

Criteria:

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

VideoEasel: A Virtual Laboratory for Statistical
Mechanics

(DFG Center
MATHEON
)

Core here:

A flexible programmable cellular automaton
implements the time evolution &
measurement tools.

=> A rich set of phenomena
based on elementary rules



Ferromagnetism and the Ising Model


Lattice Gases


Entropy and the 2nd Law of Termodynamics


Image Denoising (Geman & Geman)


Partial Differential Equations


Image Processing by Convolution Filtering

Example: The Ising Model

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

connec
-

tors

connec
-

tors

front
-
end

front
-
end

alternative

user interfaces

Software Architecture of VideoEasel

„Virtual Lab“

(simulation &

computation)

user management

Interfaces

external

Virtual

Labs

external

numerical

software

& CAS

Integration of external tools

browser/

interface

browser/

interface

browser/

interface

cooperative usage

intelligent

assistent

front
-
end

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Multiple User Interfaces

...depending on the deployment of the laboratory.

Java
-
Applet

OOrange, Maple

e.g. for demo and on
-
line experimens

e.g. for research purposes

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Modules within Virtual Laboratories


Labs focussing on physics require two types of equipment:


Objects that simulate the physical laws and that are to be measured on


Objects that simulate measurement tools

Labs as Construction Kits: Measurement tools can be attached to
“suitable” simulations

Ising,

micro
-

canonical

Ising,

bounded

Magnetization

Free Energy

Entropy

BZ
-

Reaction

Laboratory Core

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Integration into the Course Management

Course in the Content Area

Exercises in the Training Area

Asset Level (e.g Virtual Lab Area)

Matrix

Determinants

Compute

det(...)

Laplace

Gauss

Algorithm

Solve

equations

Build

Triangular

Matrix

What is

triangular?

What is a


matrix?

Elementary

Operations

requires

requires

recommends

Relations between
Exercises

Dependencies of Knowledge
Atoms

Storyboarding of
Training Units

requires…

…or requires

Kernel

Gauss

Algorithm

requires

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

The Asset Level in Virtual Laboratories:

Wizards ease the setup of the
image filter

Example: Course on Matrix
Convolution

Tutor programs
provide exercises,
validate inputs and
give hints

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Storyboards at the Asset Level:

Exercise 1

Exercise 3

Exercise 2

Exercise 4

Exercise 6

Exercise 5

Move image to the
right...

Image moves left: Recall
definition of convolution

Move image upwards

Build smoothing filter

not isotropic

Image moves somehow:
What
´
s x and y?

Adaption of the graph
according to a user profile

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Assets built from Components



Asset are reusable nodes of the exercise network.



Several copies of the same exercise are available, depending on the
target audience.



reusable evaluators classify user inputs.



Dependencies between assets are resolved automatically.

Node A

Node B

Node C

Node D

Evaluator

requires

Identical content in
varying presentation

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Assets built from Components (2)



Every asset comes with a collection of hints to be given to the
learner on request.



Evaluators assign credit points to the user and thus built
-
up a user
profile.

Node A

Node B

Node C

Evaluator

Hint 1

Hint 2

Hint 3

one credit

two credits

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Inside an Asset Node


node={



name = "Averaging by
Convolution";



explanation = "Smooth the image";



requires = "convolution" |
"filter";



provides = "average filter";



hint = "try positive weights";



audience = "Basic" | “Advanced”;




using(evaluator = "CheckFilterType"){



if (!"LowPass")

{



target = "...";



explanation = "...";



credits = 1;



}



}



# ... more goes here...


};


assignment

dependencies

target audience

evaluation and
branching

updates the user profile

hints

TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

Outlook:

Research plan:
Detect the „right branch“ in the storyboard:




Field
-
specific issues: Evaluate of the user input for
correctness, evaluate errors („hard conditions“)




Psychological issues: Describe users by psychological
variables, e.g.
Textual
vs.
Graphical

or
Holistic

vs.
Serialistic

(„soft conditions“) to select the right “audience”
for the storyboard.




Linguistic issues: Offer ways to escape from the
storyboard by parsing natural
-
language queries on the
subject.


TU Berlin


Thomas Richter

Th. Richter: Intelligent Agents for User Guidance in Virtual Labs

Online Educa Workshop

The End!