Adaptive Learning Systems

1

Adaptive Learning Systems

Associate Professor KinshukInformation Systems Department

Massey University, Private Bag 11-222Palmerston North, New ZealandTel: +64 6 350 5799 Ext 2090

Fax: +64 6 350 5725Email: [email protected]

URL: http://fims-www.massey.ac.nz/~kinshuk/

2

Introduction

• Adaptive learning systems with particular focus on cognitive skills

• Accommodation of both the ‘instuction’ and ‘construction’ of knowledge

• Design based on informed educational methodologies

3

What exactly we mean by

Adaptivity

in

Adaptive Learning Systems?

4

“Intelligence”/adaptivity

Increased user efficiency, effectiveness and satisfaction

by

Improved correspondence between learner, goal and system

characteristics

5

Need of Intelligence/adaptivity

Users generally work on their own without external support.

System is used by variety of users from all over the world.

Customised system behaviour reduces meta-learning overhead for the user and allows focus on completion of actual task.

6

Adaptable SystemsSystems that allow the user to change certain system parameters and adapt the

system behaviour accordingly.

Adaptive SystemsSystems that adapt to the users automatically based on system’s assumptions

about user needs.

7

8

How does adaptivity work? System monitors user’s action

patterns with various components of system’s interface.

Some systems support the user in the learning phase by introducing them to system operation.

Some systems draw user’s attention to unfamiliar tools.

User errors are primary candidate for automatic adaptation.

9

Levels of adaptation

Simple: “hard-wired”

Self-regulating: monitors the effects of adaptation and changes behaviour accordingly

Self-mediating: Monitors the effects of adaptation on model before putting into practice

Self-modifying: Capable of chaging representations by reasoning about the interactions

10

Problems in adaptation

User is observed by the system, actions are recorded, giving rise to data and privacy protection issues.

Social monitoring becomes possibility.

User feels being controlled by the system.

User is exposed to adaptation concept favoured by the designer of the system.

User may be distracted from the task by sudden automatic modifications.

11

Recommendation for adaptive systems Means for user to (de)activate or

limit adaptation procedure

Offering adaptation in the form of proposal

User may define specific parameters used in adaptation

Giving user information about effects of adaptation hence preventing surprises

Editable user model

12

Domain competence

And

computers

13

Constituents of Domain Competence

Know-whyKnow-how

Know-how-not Know-why-not

Know-when

Know-when-not

Know-what

logical processes

Know-about

Easier tolearn from mistakes

An example of the know-how aspect of know-when is the temporal context required for an appropriate sequence of operation

An example of the know-why aspect of know-when is the environmental and behavioural contexts required for making a decision

Action oriented and experiential

Reflection oriented and abstract

Difficult tolearn from mistakes

Trial and error

Context oriented and both experiential and abstract

Awareness oriented

14


Know-how It has an operational orientation. It is mainly action-driven and hence pre-

dominantly experiential. It is difficult to inherit it from someone

else’s experience.

Know-how-not Learning by mistakes.

Examples : Computer simulation and virtual reality

15


Know-why It has a causal orientation. It is mainly reflection-driven and therefore

based on abstraction. It can be inherited from someone else’s line

of reasoning.

Know-why-not Logical processes. Needs deeper reflection.

16


Know-when (and -where)

It has a contextual orientation.

It provides the temporal and spatial context

for both the know-how and know-why. It is

thus both action and/or reflection driven.

17


Know-about

It has an awareness orientation.

It includes above three types of knowledge in

terms of know-what.

It also contains information about the

environmental context of this knowledge.

18

Ideally, an instructional system, designed for novice users, teach all knowledge constituents.

But, know-why is difficult to handle mainly for two reasons:1. It needs natural language interaction.2. It needs use of metaphors, which are difficult to

understand for a novice user.

Know-how, on the other hand, is operational, and can be conveyed to the user more easily, even with symbolic representations.

Instruction in knowledge context

19

Traditional hypermedia based ITSs approach, in general, has been to teach the know-why aspect of knowledge with the help of explanations.

The links provide stimulus to the user to know more about a particular topic.

System works more as a friendly librarian and learning depends on the initiative of a student.

Instruction in knowledge context

20

Theoretical framework best suitable for facilitation of

cognitive skills?

Cognitive Apprentice Framework

21

Cognitive apprenticeship framework

Modelling: Learners study the task pattern of experts to develop own cognitive model

Coaching: Learners solve tasks by consulting a tutorial component of the environment

Fading: Tutorial activity is gradually reduced in line with learners’ improving performance and problem solving competence

22

Phases of Cognitive apprenticeship

1. World knowledge (initial requirement)

2. Observation of interactions among masters and peers

3. Assisting in completion of tasks done by master

4. Trying out on own by imitating

23


5. Getting feedback from master

6. Getting advise for new things on the basis of results of imitation, comparing given solution with alternatives

7. Reflection by student, resulting from master’s advice

24


8. Repetition of process from 2 to 7 Fading out guidance and feedback Active participation, exploration

and innovation come in

9. Assessment of generalisation of the tasks and concepts learnt during repetition process

25

Example system

Cognitive apprenticeship based learning environment (CABLE)

26

Environment should facilitate:

acquisition of basic domain knowledge;

application of the basic domain knowledge in non-contextual and contextual scenarios to get skills of the discipline; and

generalisation of the domain knowledge to get competence of applying it in real world situations.

CABLE objectives

27

CABLE architecture Observation - for acquisition of concepts

Simple imitation - skills acquisition through articulation of concepts

Advanced imitation - generalisation and abstraction of already acquired concepts and for acquisition of skills of applying concepts in different contexts

28

CABLE architecture Contextual observation - deeper learning after imitation process results into the identification of gaps in learner’s current understanding of the domain knowledge

Interpretation of real life problems - for acquiring competence in such narrative problems as encountered in real life situations

29

CABLE architecture

Mastery in skills - for repetitive training

Assessment - for measurement of overall progress

30

CABLETeacher generated

contextual problemsfor generalised

learning & testing

Teacher generatedcontextual problems for

strongly situatedlearning & testing

System generatedproblems - random

selection of variables

Teacher generated richnarrative problems with model

answers to simulate real lifeconditions

Descriptive text,illustrations andsolved examples

Use offine-grained

interfaces

Fine-graineddynamicfeedback

“Why ?” explanation for the system

recommended solution

“What did I do ?”diagnosticfeedback

Tools of the Trade

Assessment

Intelligent Tutoring Tools

Listen/ Observe Domain’s

concepts andtheir purpose

Interactive Learning Rehearsing/repairingmisconceptions and

missing concepts

Testing Abstract

orSingle context

Testing Multiple contexts

and/orRich narrative

Extending Greater complexityBuilding skills inthe use of tools

Learning by syntactic mapping of interfaceobjects is possible

Ensures generalisation and far transfer ofknowledge

Instruction as themain source

31

A network of inter-related variables where the whole network remains constant.

Example, partial network of 7 out of a total of 14 variables in marginal costing.

Intelligent Tutoring Tools Structure

32

Marginal costing relationships

R

VT CT

VU

Q

CU

R = VT + CTR = Q * P

P

CT = R - VTCT = Q * CU

Q = VT / VUQ = CT / CUQ = R / P

CU = CT / QCU = P - VU

VU = VT / QVU = P - CU

VT = R - CTVT = Q * VU

P = R / QP =VU + CU

33

Structure of an ITT

Inference Engine

Context based link to textual

description

User Interfacemodule

FileManagement

Input (student answer, position)Feedback

(four levels)

Knowledge Base1. Variables2. Relationships3. Tolerances

Modes- Student- Lecturer- Administrator

RandomQuestionGenerator

DynamicMessaging

System

TutoringModule

Expert Model1. Correct values2. Derivation procedure(Local expert model)

Student Model1. Student input2. Value status (filled or blank)3. Derivation procedure4. Interface preferences

Add-ons1. Calculator2. Table Interface3. Formula Interface

}Application specific

MarkerLecturer’s model answer to

any lecturer generated narrative questions

(Remote Expert Model)

34

Tutoring Strategy of an ITT

Introduction of complexity in phased manner

Corrective, elaborative and evaluative aspects of student model are used for tutoring.

Learning process is broken down to very small steps through suitable interfaces.

‘Road to London’ paradigm is adopted to eliminate the need for diagnostic, predictive and strategic aspects.

35

CABLE Demo

Future work on mental process modelling

Documents

Adaptive Learning Systems