This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Challenges and opportunities Challenges and opportunities in e-Learningin e-Learning
This work benefits from funding by the This work benefits from funding by the Romanian Ministry of Education, Research Romanian Ministry of Education, Research
and Youth and Youth (INFOSOC - CEEX 73/31.07.2006) (INFOSOC - CEEX 73/31.07.2006)
Socio-Cultural Models Socio-Cultural Models implemented through multi-agent implemented through multi-agent
architecture architecture for for
e-learninge-learning
MMain objectiveain objective:: the development of a the development of a model for the virtual education system, model for the virtual education system, student centred, that facilitates the student centred, that facilitates the learning through collaboration as a learning through collaboration as a form of social interactionform of social interaction
BackgroundBackground
Vygotsky's theory of social cognitive Vygotsky's theory of social cognitive development (1934/1986):development (1934/1986): described described learning as being embedded within learning as being embedded within social events and occurring as a child social events and occurring as a child interacts with people, objects, and interacts with people, objects, and events in the environment events in the environment
Bandura's social learning theoryBandura's social learning theory
STUDENT MODELSTUDENT MODEL
an action an action ii is defined by a set of is defined by a set of attributes A. An individual has to decide attributes A. An individual has to decide among a set among a set II of possible actions: of possible actions:
II = {i= {ikk, k=1,2,...,n} and , k=1,2,...,n} and
AA kk= {a= {akiki, i=1,2,....m} (1), i=1,2,....m} (1)
qualityquality--cost functioncost function
STUDENT MODEL – cont.STUDENT MODEL – cont.
At individual level we are considering that the At individual level we are considering that the evaluation is also influenced by two categories of evaluation is also influenced by two categories of factors: factors: beliefsbeliefs (cognitive) and (cognitive) and affectsaffects (emotive). (emotive).
iik k = ∑= ∑ii B Bii E Eii (2) (2)wherewhere
iikk is an action, k=1,..,n is an action, k=1,..,n
BBii is the belief that is the belief that iikk posses the attribute posses the attribute aakiki, , i=1,..,mi=1,..,m
EEii is the evaluation or utility (desirability) of is the evaluation or utility (desirability) of attribute attribute aakiki, i=1,..,m, i=1,..,m
∑∑ii BBii E Eii
Beliefs about Beliefs about attributes Battributes Bii
Evaluation of Evaluation of attributes Eattributes Eii
Intervention/action ik
STUDENT MODEL – cont.STUDENT MODEL – cont.
Student attitude toward an act, αStudent attitude toward an act, αkk,, is the sum of the is the sum of the student's belief strength in the consequences student's belief strength in the consequences resulting from performing a certain action (taking a resulting from performing a certain action (taking a certain decision)certain decision) weightedweighted by the evaluation of an by the evaluation of an anticipated outcome (positive benefit or avoidance of anticipated outcome (positive benefit or avoidance of a negative consequence):a negative consequence):
ααkk = ∑= ∑ii ββkiki ε εkiki (3) (3)wherewhere
ααkk is the attitude toward an action i is the attitude toward an action ikk, k=1,..,n, k=1,..,n ββkiki is the belief that performing is the belief that performing iikk will lead to an will lead to an anticipated outcome anticipated outcome ii, i=1,..,m, i=1,..,m εεkiki is the evaluation or utility (desirability) of the is the evaluation or utility (desirability) of the outcome outcome ii, i=1,..,m, i=1,..,m
STUDENT MODEL – cont.STUDENT MODEL – cont.
The influence of the colleagues from the learning environment The influence of the colleagues from the learning environment can be modeled by introducing the subjective norm: can be modeled by introducing the subjective norm:
SNSN = ∑= ∑jj NB NBkjkj MC MCkj kj (4) (4)wherewhere
SN is the SN is the subjective normsubjective norm - the motivation toward an action - the motivation toward an action ik, k=1,..,n, as determined by the influence of the groupik, k=1,..,n, as determined by the influence of the group
NBNBkjkj is the is the normative beliefnormative belief that people from the group (j) that people from the group (j) expect an individual to perform an action expect an individual to perform an action iikk will lead to will lead to jj, , j=1,..,nj=1,..,n
MCMCkjkj is the is the motivation to complymotivation to comply with the expectation of with the expectation of the group (j) the group (j) jj, i=1,..,n, i=1,..,n
The theory of reasoned actionThe theory of reasoned action
combining the attitude toward an act and the combining the attitude toward an act and the subjective norm:subjective norm:
DB = f[(BI) = f (αDB = f[(BI) = f (αkk)w)w11 + (SN)w + (SN)w22] (5)] (5)wherewhere
DB is the decisional behaviorDB is the decisional behaviorBI is the behavioral intentionBI is the behavioral intention
ααkk is the attitude toward performing the action is the attitude toward performing the action iikkSN is the subjective normSN is the subjective norm
ww11 and w and w22 are evaluation weights determined are evaluation weights determined empiricallyempirically
Student cognitive model Student cognitive model
Vygotsky Vygotsky zone of proximal developmentzone of proximal development (ZPD) (ZPD)
four learning stages: four learning stages: – assistance provided by more capable others assistance provided by more capable others
(coaches, experts, teachers); (coaches, experts, teachers); – self assistance; self assistance; – internalization automatization (fossilization); and internalization automatization (fossilization); and – de-automatization: recursiveness through prior de-automatization: recursiveness through prior
stages. stages.
Vygotsky's theory also claimsVygotsky's theory also claims
"that instruction is most efficient when students engage in "that instruction is most efficient when students engage in
activities within a supportive learning environment and activities within a supportive learning environment and
when they receive appropriate guidance that is mediated by tools"when they receive appropriate guidance that is mediated by tools"
TUTOR - modelTUTOR - model
The The TUTORTUTOR assistant evaluates the educational assistant evaluates the educational objectives of the student and recommends her/him objectives of the student and recommends her/him some kind of activities. The decisions are based on the some kind of activities. The decisions are based on the knowledge of the students’ cognitive profile (which knowledge of the students’ cognitive profile (which takes into account the social component). The takes into account the social component). The TUTORTUTOR agent interacts with the personal assistant of the agent interacts with the personal assistant of the student, with the mediating agent and with the social student, with the mediating agent and with the social agentified environment. As the system is conceived, the agentified environment. As the system is conceived, the accent is put on collaboration activities between accent is put on collaboration activities between students, which consist in knowledge exchange, students, which consist in knowledge exchange, realization of common projects, tasks’ negotiation, realization of common projects, tasks’ negotiation, sharing resources, common effort for the understanding sharing resources, common effort for the understanding of a subject, problem-solving in-group. The of a subject, problem-solving in-group. The TUTORTUTOR is is mainly evolving in a network populated with learning mainly evolving in a network populated with learning objects. objects.
Ant social behaviourAnt social behaviour
Pierre-Paul GrassPierre-Paul Grass –– stigmergy stigmergy (indirect, non-(indirect, non-symbolic form of communication,medisated by symbolic form of communication,medisated by the environment; local information)the environment; local information)
Deneubourg et al. - studies on pheromone Deneubourg et al. - studies on pheromone laying and general ant behavior (double bridge laying and general ant behavior (double bridge experiment)experiment)
→ → main source of inspiration for the main source of inspiration for the development of ant colony optimization (ACO)development of ant colony optimization (ACO)
ACOACO
In ACO, a number of artificial ants build solution In ACO, a number of artificial ants build solution to an optimization problem and exchange to an optimization problem and exchange information on their quality via a communication information on their quality via a communication scheme that is reminiscent of the one adopted by scheme that is reminiscent of the one adopted by real antsreal ants.(Dorigo et al., 2006).(Dorigo et al., 2006)
ACO algorithms use a population of ACO algorithms use a population of agents-agents-artificial antsartificial ants in order to build a solution to a in order to build a solution to a discrete optimisation problem. Solutions’ discrete optimisation problem. Solutions’ information (individual connections of the information (individual connections of the problem) are kept in a global memory – the problem) are kept in a global memory – the pheromone mappingpheromone mapping. Specific heuristics can be . Specific heuristics can be considered on a-priori information.considered on a-priori information.
3 processes3 processes
ants generation and activity, ants generation and activity, pheromone trail evaporation and pheromone trail evaporation and daemon actions. daemon actions.
Virtual learning environment: Virtual learning environment: a set a set GG of students’ teams and of students’ teams and a set a set AA of group work activities of group work activities consisting of several tasks.consisting of several tasks.LetLetaajj AA with j=1,…, with j=1,…,J J be an activity from the set be an activity from the set AA ; ;
(1) (1)ggii GG with i= 1,… with i= 1,…II be a team from the set be a team from the set G,G, and and
(2) (2)tktkjiji TK , TK , j=1,…,j=1,…,JJ and i= 1,… and i= 1,…I I
(3) (3) be a task of the activity abe a task of the activity ajj that has to be that has to be
performed by the team gperformed by the team gjj..
activityactivity a aj j :an ordered sequence of tasks :an ordered sequence of tasks from a set TK = { tkjifrom a set TK = { tkji }.}.
tasktask tktkjiji : to be performed by the team : to be performed by the team ggii in a number in a number ddjiji of time units. of time units.
N = │TK│- N = │TK│- total number of tasks. total number of tasks. Goal: Goal: to assign tasks to time intervals to assign tasks to time intervals
in such a way that no two activities are in such a way that no two activities are performed at the same time by the same performed at the same time by the same team and that the maximum completion team and that the maximum completion time of all tasks is minimized. time of all tasks is minimized.
A team gA team gii G G is supposed to haveis supposed to have M M members members (student-agents):(student-agents):
ggii = g = gimim with m=1,.., M (4) with m=1,.., M (4)
- all teams have the same number of membersall teams have the same number of members - there are task-classesthere are task-classes - the potential assignment of one student to a the potential assignment of one student to a
specific task is based on the concept of specific task is based on the concept of response threshold combined with a function response threshold combined with a function of the background knowledge of the student of the background knowledge of the student and her decision making behaviour.and her decision making behaviour.
A student-agent, gA student-agent, gimim , is attracted to a task t , is attracted to a task tjiji with a probability with a probability PP depending on her depending on her background knowledgebackground knowledge, , beliefsbeliefs (cognitive), (cognitive), affectsaffects (emotive) and (emotive) and intentions intentions and on and on response threshold response threshold θθimj imj ::
PP θθ imjimj (s (simjimj) = s) = simjimj 22 / / (s (simjimj22 + + θθ imjimj
22) (5)) (5)
wherewhere s simj imj = = f f (beliefs, affects, intention, qualification-(beliefs, affects, intention, qualification-
The response threshold The response threshold θθ imjimj of the ant-agent of the ant-agent ggimim to the to the task tktask tkji ji is decreased when the agent has performed the is decreased when the agent has performed the task tktask tkjiji ; at the same time, thresholds for the other ; at the same time, thresholds for the other tasks are increasing proportional to the time t to tasks are increasing proportional to the time t to perform the task.perform the task. θθ imj imj
where:where:xxijij is the time spent by the agent i for the task j, is the time spent by the agent i for the task j,ξξ is a learning coefficient , and is a learning coefficient , andφφ is a forgetting coefficient. is a forgetting coefficient.
If a task is performed, the response threshold is If a task is performed, the response threshold is
decreased with a quantity depending of the learning decreased with a quantity depending of the learning coefficient, in the opposite situation; the threshold is coefficient, in the opposite situation; the threshold is increased with a function of the forgetting coefficient.increased with a function of the forgetting coefficient.
The state of an ant-agent is evolving from The state of an ant-agent is evolving from “active” – performing a task, to “inactive” –“active” – performing a task, to “inactive” –idle, and vice versa. An inactive agent starts to idle, and vice versa. An inactive agent starts to perform a task with a probability perform a task with a probability PP given by given by (5). An active agent completes the task, or (5). An active agent completes the task, or abandons it, with a probability abandons it, with a probability pp per time unit : per time unit :
pp = = probability probability (state = “active” → state = “inactive”) (state = “active” → state = “inactive”) (7) (7)
1/1/pp is the average time spent by an agent in is the average time spent by an agent in task performing before giving up the task.task performing before giving up the task.
The problem is represented as a weighted graph The problem is represented as a weighted graph Q = Q = (TK’(TK’,, L) L), where , where TK’= TK {tkTK’= TK {tk00}} and and LL is the set of is the set of edges that connect node edges that connect node tt00 with the first task of each with the first task of each activity. The vertexes of T are completely connected, activity. The vertexes of T are completely connected, with exception of the nodes of the tasks from the same with exception of the nodes of the tasks from the same activity that are connected sequentially, each such node activity that are connected sequentially, each such node being linked only to its direct successor. being linked only to its direct successor.
There are N (N-1)/2 + │A│ edges. Each edge (k, l) is There are N (N-1)/2 + │A│ edges. Each edge (k, l) is weighted by two numbers: weighted by two numbers: ττlklk - the - the pheromone levelpheromone level (trail level) and (trail level) and ηη klkl – the so called – the so called visibilityvisibility and and represents the desirability of a transition from node k to represents the desirability of a transition from node k to node l. Each ant-agent has associated a data structure – node l. Each ant-agent has associated a data structure – the the tabu-list, tabu-list, that memorises the tasks of an activity that memorises the tasks of an activity that have been performed at the time moment t.that have been performed at the time moment t.
A A transition probability functiontransition probability function from node from node i i to node to node jj for an ant-agent (a-a) for an ant-agent (a-a)kk was defined was defined as:as:
(8)
where allowedk = (N – tabuk)
tabuk is a vector that changes dynamically and contains the tabu-list of the kth ant.
ά and β are parameters that are used to control the relative importance of pheromone level and visibility
NR - number of potential active agents, NR - number of potential active agents,
NRNRactact - number of active agents at the time t, - number of active agents at the time t,
The variation of the attraction of a task The variation of the attraction of a task (pheromone deposit) in a discrete time (pheromone deposit) in a discrete time situation:situation:
ssimjimj (t+1) = s (t+1) = simjimj (t) + (t) + ββ - ( - (άάNRNRactact) / NR (9)) / NR (9)
The order in which the nodes are visited by The order in which the nodes are visited by each ant-agent specifies the proposed each ant-agent specifies the proposed solution.solution.
ConclusionConclusion
““We shape our dwellings and We shape our dwellings and afterwards our dwellings shape our afterwards our dwellings shape our lives”.lives”.