Personalized Interactive Tutoring in - PowerPoint Presentation

Slide1

Personalized Interactive Tutoring in Chess

Department of Computational and Cognitive NetworksAcademy of Sciences of ArmeniaInstitute for Informatics and Automation Problems

Edward

Pogossian

epogossi@aua.am

Sedrak

Grigoryan

addressforsd@gmail.comSlide2

1.Why Tutoring

andwhat we did Slide3

Actuality:

- basic knowledge is passed to descendents only first hand - students learn in different ways - unordinary students: require personalized approach Premises: - advances in computer sciences

=>

make possible

personalized interactive

tutoring

and

examining

- certain

types of exams

can be interpreted as

game problems

Questions

How to

provide

experts with adequate computer

tutoring tools?

How to examine

the

acquisition of knowledge

?Slide4

What we

did:

1. We provide

experts

with a computer

tutoring

tool based on

s

olvers of chess-like games

2. The adequacy of our models rely on consistency of knowledge presentation and processing with ones in English and experiments in tutoring chess endgames

Slide5

3. E

ffectiveness of learning by tool is measured in scales and by methodology consistent with ones of experts 4. Solvers of chess like games can be a base for effective tutoring

5.

T

utoring tools have to be developed in close cooperation of all parties involved in education and cognitive modelingSlide6

How we did?

1. RGT Problems

2.

RGT Solvers

3

.

Modeling

Tutoring

4. Adequacy of

Models

of Tutoring

5.ConclusionSlide7

1.

RGT ProblemsSlide8

Unsolved problems

Solved problems

կ

Unsolved combinatorial problems

RGT

Interpreting unsolved problems by solved ones

Solving by Modeling Human

Approaches

TutoringSlide9

There are Interacting Actors

2.

Actors may perform actions

Action1

Action2

1.

3.

There are specified types of situations

Situation1

Situation2

Some situations are selected as Goals

4.

Situation1

Situation2

Actors’ Actions transform Situations

5.

Sit1

Sit2

Sit3

Action2

Action1

Game Tree

RGT Problems Meet the Following RequirementsSlide10

RGT

Class of Problems

Chess

Management

Intrusion Protection

Defense of Military Units

Anomalies detection in computations

Problems of Testing

Kernel

RGTSlide11

2. RGT SolversSlide12

Strategy Search in RGT Solvers

Input Situation

G1/B

G1/C

Actions

G1/A

Goals

Goal1Slide13

Controller

Store of

Abstracts, Goals, Plans

Store of

T-Prints

Graph of

Abstracts

Abstract

Matcher

GUI

Abstracts Acquirer

Matching Visualizer

T-Prints

Perceiver

Problem Manager

A1

A2

A3

A5

A41

A6

Abstracts

Sub1

Sub2

Classifier

Method,

[0/1], Name

List of Attributes

T-Print

PPIT

CPMU

GP

RHP

Acquirer

RGT Solver

s

Actions by Moves

Knowledge RevealerSlide14

3. Modeling TutoringSlide15

CheckMate

King under check

King can’t escape

King has no defense

Level i

Level i+1

How Experts Are Tutoring ?

1. Student

has certain level of knowledge (e.g. knows chess basic rules)

2. Teaches

for unknown chess concepts required for the solution

3. Teaches

for the plan to play (e.g. Push king to an edge, make opposition and put mate)Slide16

Tutoring by RGT expert

RGT Expert

Personalized

Interactive

Tutoring Environment Based on RGT Solver

Adequate to

RGT Solver

RGT Knowledge models Adequate to Expert

Strategy Search Algorithms Adequate to Expert ApproachSlide17

Tutoring Environment

Tutoring for Chess ConceptsTutoring for Strategies

1. Explanation of Chess Concepts

2. Providing examples of chess concepts

1. Explanation of Plans and Goals

2. Providing examples of performances of plans

3. Testing of understanding

Student has background of understanding chess, figures, colors (black and white), board, movesSlide18

RGT Solvers in Tutoring

RGT Solver

Tutoring Protocol

Generation of

Example

s

Generation of Testing Situations

Explanations of Classifiers and Strategies

Tutoring Environment

Feedback provision mechanisms to identify bad described RGT knowledge (for improvement purposes)

Interfaces for

Integration of RGT problems

Chess Tutoring

I

nterface

Future Steps

Completed

Partially completed

Testing of RGT knowledge

Tool

for measuring the progress

of

student

sSlide19

CheckMate

King under check

King can’t escape

King has no defense

Field under check

King

Field under check of Knight

Field under check of Knight1

Figure

Field

Figure Type

Figure Color

X

Y

White or Black

King Type

Not empty type

Chess concepts explanation

1. Different levels of

explanationsSlide20

RGT Solvers provide:

- Models of RGT knowledge

-Strategy search algorithms

-Tutoring protocols

Tutoring is

Personalized

Interactive

level by level

explanation,

testing

, feedback provision and correction,

assessment of the progress of students.Slide21

Explanation

of plans and goals

Plan1

Goal2

Goal4

Goal1

Goal2

Precondition

Postcondition

Evaluator

Abstract1

Abstract2Slide22

Plan1

Goal2

Goal4

Goal1

Goal2

Precondition

Postcondition

Evaluator

Goal1

Precondition

Postcondition

Evaluator

Actions

Providing an example of performances of plansSlide23

RGT Solver

Plan

Action 1

Action 2

Action 3

Plan

Action 1

Action 4

Action 3

Correct

Wrong, Explain

Correct

Testing of AcquisitionSlide24

4.Adequacy of

Models of TutoringSlide25

Knowledge-based

Solvers have Effectiveness and Efficiency (EE) comparable

with

experts

minimax

Solvers

provide

the

idea of max

Effectivenes, but not acceptable joint EEminimax Solvers with parametric evaluation functions

Search by

minimax, parametricevaluation functionKnowledge Based Solvers

Solving by Modeling HumanApproaches, Expert Systems

Botvinnik

, Pitrat, Wilkins: Parametric methods are not adequate for combinatorial problemsSlide26

Categories of English Verbs

“Have, Be, Do” (HBD) knowledge presentation in English and in the model are consistent

Be,

Exist,…

English

Verbs

Have,

Possess,

Own,…

DoSlide27

HBD

model is consistent with OOP

1.

Abstract Name

2

.

Has

attributes

3.

Does

actions

4.

Is

inherited

from another abstractSlide28

Advantages of HBD Models

PropertyOOPOnt.

Pr.S

.

HBD

Represent different type of knowledge

+

-

-

+

Opacity

++-+Reuse++

-+Polymorphism+-

-+Inheritance+

+-+Matching data to the entities

(rules, classes etc.)--++

Dynamically change class hierarchies

-

-

-

+

D

ynamically generate

/integrate new

entities

-

+

+

+Slide29

RGT Solvers are able to process complex knowledge

in solving RGT Problems

Botvinnik

suggested tests for measuring the program’s quality: the

Reti

and

Nodareishvili

chess etudes

Personalized Planning and Integrated Testing (PPIT) 2007

Reti

etude

:

draw

Nadareishvilli

etude:

winning

Slide30

By exhaustive search Nadareishvili

etude can be solved only with the depth of 36

in

the game

tree search while experts and RGT Solver solve it analyzing about 500 positions Slide31

Tutoring Rock vs. King

Explanation of the winning strategy in Rock against King endgames:Put mateAvoid stalemateEscape rook from attackPush king to the edge (without putting rook under attack)

Make a waiting move when

preOpposition

appears

Bring white king closer to the opponent

king

Slide32

The

Plan of Rook vs King :

Put

mate

Avoid stalemate

Escape rook from attack

Push king to the edge (without putting rook under attack)

Make a waiting move when

preOpposition

appears

Bring white king closer to the opponent

king Slide33

RGT Solver

Plan

1

2

3

1.

“mate” concept

3

.

“rook under attack”

1

st

step: Explanation of Goals:

2.

“stalemate” concept

4.

“edge”, “push king to the edge”

=

4

Explain

5

.

“Pre Opposition”, “waiting move” concepts

6

.

“Opposition”

5

6Slide34

3. Escape Rook from

the Attack

Rook under attack

Rook

Field under attack

Field under attack of King

Field under attack of Knight

Field under attack of King1

Field under attack of King8Slide35

4. Push King to the edge (without putting

Rook under Attack)

Edge

EdgeVertical

EdgeHoirzontal

1. King is maximal close to edge

2. King has less movesSlide36

5. Make a waiting move when pre Opposition appears

1. Rook distance is maximal by the vertical/horizontalSlide37

6. Bring white King

closer to the black King (avoid opposition)

1. Distance between kings is minimal

Oppostion

Oppostion

by vertical

Oppostion

by horizontal

Oppostion

by horizontal 1

Oppostion

by horizontal 2Slide38

RGT Solver

2

nd

step: Example of Execution of Plans:

1. Put Mate

Move R g5 selected

2. Avoid Stalemate

3. Escape rook from attack

4. Push king to the edge

Similarly next situations are processed and explained

Plan

1

2

3

=

4

5

6Slide39

RGT Solver

3

rd

step: Examining Understanding of Plans:

1. Put Mate

Selected K e3 move is

correct

2. Avoid Stalemate

3. Escape rook from attack

4. Push king to the edge

Similarly next situations are checked,

if wrong, corrected and explained as in (1,2 steps)

Plan

5. Make a waiting move

6. Bring king closer to opponent

K e

3

move is performed by student

Plan

1

2

3

=

4

5

6Slide40

Measuring Progress of Students

Knowledge-Based Solvers against Knowledge-Based Solvers

Knowledge-Based Solvers against Experts (students)

Experts (students) against Experts (students)Slide41
Slide42

Chess ratings based scales and methodlogy of the quality of RGT Solvers are developed

Strong

measurement

of

quality

of modifications of RGT

Solvers and their constituentsSlide43

5. ConclusionSlide44

1.

We provide

experts

with a computer

tutoring

tool based on the RGT Solvers

2. The adequacy of model of tutoring

to one of experts

was successfully examined

3.

Adequate scales and methodology were developed to measure the

effectiveness of tutoring4.

RGT Solvers are the base for effective models of tutoring

5. Development of effective

tutoring tools needs close cooperation of educators and

cognitive

modelers Slide45

Thank You !Slide46

4. Advances in

RGT Solutions Slide47

Confirming Adequacy of

Models of RGT Knowledge and Matching Algorithms

It was confirmed for:

Chess

Marketing

Intrusion ProtectionSlide48

Strongly specified RGT class of problems

Chess ratings based scales of the quality of RGT SolversAdvances in solving

particular RGT problems are interpretable for

RGT

class

: unified Solvers

can be constructed

Knowledge-based

Solvers can provide

EE comparable with human experts (Botvinnik

: Parametric methods are not adequate for combinatorial problems)

Knowledge consist of Strategies (regularities), Classifiers of situations, Goals and Plans

RGT Knowledge is constructive and can be simulated

Advances in RGT SolutionsSlide49

RGT Knowledge-Based Solvers overcome RGT minimax

Sovlers by EE.IGAF1

and IGAF2 RGT Solvers Based on Common Planning

vs

Minimax

Solvers

Diagram (in Intrusion Protection)

Number of nodes searched by the IGAF2 algorithm compared with the IGAF1 algorithm and the

minimaxSlide50

Single Ownship

Against Air Threats Ownship

and air threats as

actors

Situation

with

ownship

and threads in certain distance range.

Ownship

goal

: to defend,

air threats goal: to make damageActions of ownship

: A. launch a long range surface-air missile (SAM), B. shoot the medium range gun C. shoot

the short range gun.Actions for threats:

an anti-ship missile .

Slide51

Stilman

B., USASlide52

Modeling Chess Tutoring

by RGT SolverSlide53

Personalized Interactive Tutoring by RGT Solvers

Actuality of Personalized Tutoring

RGT Solver based Tutoring

Tutoring Environment:

Tutoring for classifiers

Tutoring for strategies

Measuring the progress of students

Confirmation of Adequacy of TutoringSlide54

Actuality

: - students learn in different ways - unordinary students: require personalized approach (e.g. autistic children)

Premises

:

- advances in computer sciences

=>

make possible

personalized interactive

tutoring

and testing - certain

types of exams can be interpreted as RGT problems

QuestionsHow to provide tutoring of classifiers and strategies adequate to experts ? How to examine acquisition of knowledge adequate to experts?Slide55

Tutoring by RGT expert

RGT Expert

Personalized

Interactive

Tutoring Environment Based on RGT Solver

Adequate to

RGT Solver

RGT Knowledge models Adequate to Expert

Strategy Search Algorithms Adequate to Expert ApproachSlide56

RGT Solver In Tutoring

RGT Solvers provide

Models of RGT knowledge

Strategy search algorithms

Tutoring protocols

Tutoring is

Personalized

Interactive

level by

level

explanation,

testing, feedback provision and correction,assessment of the progress of students.Slide57

RGT Solver

Tutoring Protocol

Generation of Example

Generation of Testing Situations

Explanations of Classifiers and Strategies

Tutoring Environment

Feedback provision mechanisms to identify bad described RGT knowledge (for improvement purposes)

Testing of RGT knowledge

Interfaces for

Integration of RGT problems

Chess Tutoring interface

Tools for

M

easuring the Progress of Student

RGT Solvers In TutoringSlide58

Tutoring for concepts Tutoring for strategies

Examples for knowledge

Testing of knowledgeSlide59

Conclusion

Methods and software for tutoring to chess are developed within RGT Solver. The approach gives the following advantages:

The mechanism of tutoring is personalized for each

student.

Level by

level tutoring, testing are

provided in the interactive environment.

Students’ performance measurement means are provided in the developed interface tool

.