Emergent Intelligence: A Novel Computational Intelligence Technique to

Solve Problems

Suresh Chavhan and Pallapa Venkataram

Department of Electrical Communication Engineering, Indian Institute of Science, Bangalore, India

Keywords:

Emergent Intelligence, MultiAgent System, Job Scheduling and Resource Allocation.

Abstract:

Technological advancement and increasing globalization makes humans face many problems in day to day life,

involving many possible goals and each goal is associated with multiple possible actions, each associated with

many different dynamic and uncertain consequences. In real systems, the message passing mechanisms and

few computational intelligence techniques (like Swarm intelligence, Multiagent System, etc.) hinder mutual

cooperation and coordination of agents while solving problems in an uncertain environment, even though they

are highly efﬁcient and sophisticated. Therefore, in this paper, we propose an Emergent Intelligence technique

(EIT) based problem solving. The EIT is collective intelligence of group of agents, which is an extension of

multiagent system (MAS). Unlike MAS, the EIT provides independent decision making for a single task by the

multiple agents with mutual coordination and cooperation. It is very useful to solve the complex and dynamic

problems in uncertain environments. In this paper, we discuss EIT functioning, beneﬁts, comparisons, and also

illustration of two problems: (1) resource allocation and (2) job scheduling. Each problem is categorically

analyzed and solved step by step using EIT. We measure performance of the technique by considering real

time situations, and results are compared and shown the importance of EIT over MAS.

1 INTRODUCTION

Problem solving is the process of working through

details of a problem to reach one or more solu-

tions. Problems have some typical characteristics,

such as complexity (large number of items, interrela-

tions and decisions), dynamics (time considerations),

in-transparency (lack of clarity of situation) and poly-

tely (multiple goals). Problem solving may include

mathematical or systematic operations and can be a

gauge of an individual’s critical thinking skills. The

problems are solved using problem solving methods

includes Swarm intelligence, Agent based system,

heuristic algorithms, etc.

Agent-based systems use the set of intelligent

agents for complex problem solving and decision

making. Software agent systems allow a single agent

intelligently interacts and collects relevant informa-

tion to make a decision. In other words an agent

is an authority to take decisions while solving prob-

lems, like conﬂict problems (Jacak, et.al., 2007), E-

learning (Sun, S., et.al., 2007), medical (Fenza, G.,

et.al., 2012), process automation (Pakonen, A., et.al.,

2007), image analysis (Bell, D.A., et.al., 2007), etc.

Swarm intelligence based system solves asset man-

agement problem (Reynolds, Joshna, et. al., 2015),

real time allocation problem (Johansson, F.,, et. al.,

2010), scheduling problem (Kalyan V., et. al., 2004),

evacuation problem (Guizzi, et. al., 2015), etc. These

methods take indepedent local decisions, individual

interactions, self-aware, provides local view and they

concern managing the agents behavior, like migra-

tion, interaction, mobility, etc., and these behavior

makes less efﬁcient and may lead to delays and less

accurate in decision making.

Hence, we have developed a technique called EI

where multiple agents would take partial or full de-

cisions on a task by using the available and histori-

cal information and the decisions are communicated

to the main task originator. An EI is not attributed

to an individual agent, but it is a global outcome of

group of agents coordination, cooperation, collabora-

tion and negotiation. The Emergent Intelligence is a

main concern of potential users when we try to uti-

lize agent technology for transportation planning, au-

tonomous systems, development of smart city, smart

grid computing, etc. In addition, The EI takes collab-

orative and collective dynamic decisions, group inter-

actions, group aware, provides global view and also

it concerns management of agents behavior and man-

Chavhan, S. and Venkataram, P.

Emergent Intelligence: A Novel Computational Intelligence Technique to Solve Problems.

DOI: 10.5220/0007244100930102

In Proceedings of the 11th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2019), pages 93-102

ISBN: 978-989-758-350-6

agement of information, such as resources, jobs, ac-

tivities, tasks, etc. In literature (Rqevski, G., et. al.,

2007), (W.D. Hillis, 1988), (T. Wolf, 2007), (Z. Li,

et. al., 2006), and (P. J. Angeline, 1994) exhibits the

guidelines on how to exploit Emergent Intelligence

and Self-organization in multiagent systems to solve

problems. But it is not clear, how EI technique can be

used for problem solving in uncertain and dynamic

environment.

In this paper, we developed and presented a

novel technique, i.e., Emergent Intelligence Tech-

nique (EIT), which is an extension of multi-agent

system (MAS). We have demonstrated an EIT based

solution by collaborating, coordinating, cooperating

and negotiating the problem solving agents. The

proposed technique solves the problem by executing

6-phases sequentially: (1) Analyzes given problem,

makes clear problem statement and identiﬁes all pos-

sible tasks, subtasks, inputs and outputs; (2) Builds an

EI-network for the given problem; (3) Estimates pos-

sible practical solutions for each sub-task; (4) Creates

action plans; (5) Implements all action plans and (6)

Whenever any changes in input repeats phases from

1-5. In this paper, we have illustrated resource allo-

cation and job scheduling problems, each problem is

categorically analyzed and solved step by step by us-

ing EIT.

The rest of the paper is organized as follows. Sec-

tion 2 presents the literature survey. Section 3 dis-

cusses the EI technique. Section 4 presents the re-

source allocation problem using EI technique. In

Section 5 job scheduling problem solving using EI

technique is discussed. Section 6 case study is de-

scribed; Section 7 simulation and analysis results are

discussed; and conclusions are drawn in Section 8.

2 LITERATURE REVIEW

In this section, we discuss problem solving using the

intelligent systems with swarm intelligence, multia-

gent systems, and emergent intelligence.

2.1 MAS based Problem Solving

Authors in (Jacak, et.al., 2007) solved conﬂict prob-

lems by coordinating and negotiating multiagent sys-

tems actions to achieve a common goal. An agent

consists of many autonomous components in order to

perceive and react to its environment, plan and ex-

ecute an action, and intelligently it negotiates with

other agents. In (Gundersen, et.al., 2005) authors

have developed a prototype of multi-agent system

based approach to construct plans based on avail-

able resources in the environment, which dynam-

ically plans and solves assigned problems. Also,

MASs have been used to solve problems, such as E-

Learning (Sun, S., et.al., 2007), medical (Fenza, G.,

et.al., 2012), process automation (Pakonen, A., et.al.,

2007), image analysis (Bell, D.A., et.al., 2007). These

kind of systems have traditional beneﬁts of concur-

rent and distributed problem solving strategies (Bala

M., 2008). Novice users suffer from their incapability

to combine individual statements and constructs re-

lated to ﬂowchart, HIPO chart, IPO chart, algorithm,

etc., into valid programs () and (Aris., T.N., 2012).

In (Johansson, F.,, et. al., 2010) authors have focused

on investigating the possibility to use multi-agent sys-

tems as a new agent model for computational problem

solving which is utilized by visual programming as

the mode of programming to make it easier for novice

programmers.

2.2 SI based Problem Solving

Particle swarm optimization is especially useful for

rapid optimization of problem involving multiple ob-

jectives and constraints in dynamic environments.

Work in (Johansson, F., et. al., 2010) particle swarm

optimization has applied to real time allocation prob-

lems and discussed the allocation of weapons for

defensive purposes. Authors in (Reynolds, Joshna,

et. al., 2015) swarm intelligence is used for the au-

tonomous asset management problem in electronic

warfare. The particle swarm optimization speed pro-

vides fast optimization of frequency allocations for

receivers and jammers in highly complex and dy-

namic environments. In (Kalyan V., et. al., 2004)

authors have presented a swarm intelligence based

approach for optimal scheduling problems in sen-

sor networks. Authors have developed a methodol-

ogy and cost function to solve the graph partitioning

problem. The swarm intelligence algorithm solves

the problem and emerges with an optimal schedule.

Work in (Guizzi, et. al., 2015) authors have dis-

cussed the swarm intelligence based solutions to evac-

uation problems. Authors have determined the opti-

mum path during evacuation process by using swarm

intelligence’s algorithms (both ant colony and particle

swarm optimization).

2.3 EI based Problem Solving

Authors in (Rzevski, G., et., al., 2007) described

scheduler behavior using emergent intelligence in

multi-agent systems for not only transportation do-

main and all other logistics applications. Research in

ICAART 2019 - 11th International Conference on Agents and Artiﬁcial Intelligence

(T.Wolf, 2007) explained how self-organizing emer-

gent behavior exhibits in multiagent systems. Authors

in (Li, et., al., 2006) carried out the survey of emer-

gent behavior and its impacts on agent-based systems.

Authors in (P.J. Angelone, 1994) demonstrated and

provided the guidelines on how to exploit Emergent

Intelligence to extend the problem solving capabili-

ties.

3 EMERGENT INTELLIGENCE

TECHNIQUE

EIT is collective intelligence of a group of agents.

The EIT is an extension of MAS where agents group

activities and individual independent decision making

are incorporated. The group of agents cooperatively

(whenever agents are compatible towards complet-

ing tasks execution), coordinately (whenever there is

need of agents joint actions for executing tasks) and

collaboratively (whenever agent has insufﬁcient abil-

ity to carry out tasks) interacts among themselves to

provide dynamic independent decisions for the given

problems. The EIT can be used for executing inde-

pendent tasks, which could be parallelly executed and

provides partial (or complete) solution.

Figure 1: Scenario of 3 nodes network with task and its

sub-tasks.

The EIT is illustrated as follows: Let us consider

the task t

in 3-nodes network as shown in Figure 1.

As per the objectives of the task, it is possible to split

the task t

into 3-subtasks (st

, st

and st

) which

can be independently solved by 3 agents. Since the

task is initiated at node A where EIT is deployed in it,

which creates 3-agents A1, A2 and A3 and migrates

them to all 3-nodes A, B and C, respectively. These

agents independently solve the task t

by considering

all relevant information available locally and globally.

Finally, the decision is communicated to the main task

solving agent A1 as per the following equation

D(t

) = D(st

) + D(st

) (1)

where D(st

), D(st

) and D(st

) are the partial or

full decisions taken at nodes A, B and C, respectively.

The same problem would have been solved by the

MAS exclusively at node A by collecting relevant in-

formation from nodes B and C.

The EI technique is elaboratively discussed in the

Section 3.2.

In the literature, we can observe that some of

the attempts made to use EIT for solving problems

(Rzevski, G., et., al., 2007, W.D. Hillis, 1988, T.Wolf,

2007, Z. Li, et., al., 2006, P.J. Angelone, 1994). But

it is not clear, how this technique can be used for the

problem solving in an uncertain domain.

3.1 Deﬁnitions

In this section, we present some of the deﬁnitions use-

ful to discuss the proposed technique.

1. Task and its subtasks: Task can be decomposed into

independent subtasks. These sub-tasks are assigned

to agents and they execute and produce results. These

results are passed to the main task originator agent

and it takes ﬁnal cumulative decision.

2. Cluster: A cluster is logically partitioned depend-

ing upon the nodes degree of connectivitygreater than

or equal to 2. The creation of clusters in the given

problem network is given as follows: (i) Select a node

(where problem need to be solved) from the given

problem network, (ii) Agent at the selected node ﬁnds

the degree of connectivity (i.e., all 1-hop away nodes),

(iii) Agents at these 1-hop away nodes calculate their

own degree of connectivity, if they have more than 2

then remove from the list of the selected nodes, (iv)

End nodes or nodes which are having only one degree

of connectivity are used to form the cluster at the se-

lected node, and (v) Nodes which are having multiple

degree of connectivity will also follow the above steps

to form their respective cluster. Algorithm 1 explains

the procedure to ﬁnd the degree of nodes, cluster head

nodes, end nodes, and forms the clusters of the given

network.

Algorithm 1:

Begin

Let n be the total number of nodes in the given problem network

Select a node (where problem need to be solved) from n

/*Identiﬁcation of degree of nodes in the given problem network*/

Initialize D(i) = 0 (degree of node i).

for i = 1 to n do

for j = i+ 1 to n do

if node i is 1-hop neighbor of j then

D(i) = D(i) + 1

end if

Degree of connectivity of node i is Deg

= D(i)

end for

D(i) = 0

end for

/* Identiﬁcation of highest degree of nodes*/

Initialize m = 0

for i = 1 to n do

for j = i+ 1 to n do

m = max(Deg

, max(m, Deg

))

end for

Emergent Intelligence: A Novel Computational Intelligence Technique to Solve Problems

Highest degree of node is m

end for

/*Formation of clusters*/

for i = 1 to n do

if Deg

≥ 2 then

Node i is cluster head

for j = 1 to Deg

if degree of node j is < 2 then

Node j is the end node.

end if

Cluster i is formed with all the 1-hop away end nodes of node i

end for

end if

end for

End

3.EI-Network: It is formed by developing static agent

at each of cluster heads. These agents are connected

as per the requirements in the problem. The EI-

network is formed as follows: (i) EIT running on a

cluster head node (where problem need to be solved)

creates an EI-network, (ii) EI-network consists of

cluster head nodes and their connectivity to solve the

given problem, (iii) EIT creates and sends agents to

cluster head nodes in the EI-network, and (iv) In ev-

ery cluster the cluster head interacts and shares the an-

alyzed information. For example Figure 2 shows the

EI-network with having 5-cluster head nodes, where

EIT is running on node A. EI-network formation is

explained in the Algorithm 2.

Figure 2: EI-network.

Algorithm 2:

Begin

Let A be the node where problem need be solved in the given problem

network

Algorithm 1 provides the number of cluster head nodes (N

EI-technique runs on a node A

for A

= 1 to A

−1

Cluster head node A makes connection with cluster head node A

end for

for A

= 1 to A

− 1 do

Cluster head node A creates agent and sends to cluster head node A

Migrated agent interacts with cluster head node A

and shares the ana-

lyzed information of the given problem

end for

End

3.2 EI-Technique Functioning

Any given problem in terms of goals and barriers can

deﬁne many problems and splitting a big problem into

more manageable sub-problems. The EI-technique

solves a given problem in following six phases:

(P1) Analyze the problem, make clear problem state-

ment and identify all inputs and outputs

(i) It analyzes the given network, available and re-

quired information, (ii) It identiﬁes the number of in-

puts, tasks/subtasks and outputs, and (iii) It formu-

lates the problem statement using the available and

required information of the given problem along with

the constraints, such as activity and history database

information.

(P2) Build the EI-network for the given problem:

(i) Creation of clusters, which is explained in Sec-

tion 3.1(2) and (ii) EI-network formation, which is

explained in Section 3.1(3).

(P3) Estimate the practical solution for each sub-task:

(i) Each sub-task is assigned to a cluster-head in the

built EI-network, (ii) EIT makes the cluster head

node’s agents to cooperate, coordinate, collaborate

and negotiate among them to collect the resources

needed for each sub-task, and (iii) Practical solution

is estimated, of each sub-task, by referring to the ac-

tivity and resource utilization database along with the

available and collected resources.

(P4) Create an action plan:

(i) Every agents parallelly refers to history database

to estimate accurate requirements of each of the node,

(ii) In each cluster, the cluster head node’s agent col-

lects the relevant excess resources from nodes in the

cluster, and (iii) In each cluster, cluster head node’s

agent parallelly allocates the resources.

(P5) Implementation of an action plan:

(i) Interaction among cluster head node and other

nodes in the cluster, (ii) Accurate estimation of re-

quirements using history database, (iii) Excess rel-

evant resources collection from the one-hop away

nodes, and (iv) Each cluster head node’s agent par-

allelly allocates the resources.

(P6) Whenever a change in input repeat Phases from

1-5: In this phase, if any change in input or change

in external environment condition it makes to repeat

Phases from 1-5 again.

3.3 Beneﬁts of EI Technique

Beneﬁts of EIT over multiagent system (MAS) in ran-

domness and uncertain problem domains are as fol-

lows:

(1) In EIT, distributed agents take partial or full deci-

sions, whereas in MAS, decisions are unpredictable,

(2) A problem is being automatically decomposed

among distributed agents in EIT, whereas in MAS,

each agent chooses to whether to participate,

(3) Task and its sub-tasks are distributed and decen-

tralized action selection are done in EIT, whereas in

MAS, agents keep their own copies a task/sub-task

that might include others,

ICAART 2019 - 11th International Conference on Agents and Artiﬁcial Intelligence

(4) In EIT, distributed agents take the beneﬁts of lat-

est resource available, history database and also new

nodes coming into picture during execution of tasks,

whereas in MAS, takes the beneﬁts of only available

resources during execution of tasks,

(5) EIT reduces complexity (space and time) due to

the concise and natural modeling of problem domain:

(i) In MAS, takes O(n × m) time complexity and

O(n+ m) space complexity, where n is the total num-

ber of agents (or nodes) and m total number of tasks

need to be executed. (ii) In EIT, takes O(n × log(m))

time complexity and O(n+log(m)) space complexity.

(6) In EIT, brings autonomy, ﬂexibility, adaptiveness,

robustness, self-organization and evolution, whereas

in MAS, brings only autonomy,adaptiveness and ro-

bustness and

(7) EIT replenishes environment by creating au-

tonomous regenerating feedback loop by interaction

among group of agents.

3.4 Comparisons of Computational

Intelligence Techniques

In this subsection, we present comparisons among

Computational Intelligence (CI) techniques, that is

MAS, Swarm Intelligence (SI) and EIT. In MAS ap-

proach agents coordinate their knowledge and activ-

ities and reason the processes of coordination au-

tonomously. SI is used to model simple behaviors of

individuals, and local interactions with environment

and neighboring individuals, in order to obtain more

complex behaviors that can be used to solve com-

plex problems. Table 1 shows the comparisonsamong

MAS, SI and EIT.

4 EXAMPLE 1. RESOURCE

ALLOCATION PROBLEM

SOLVING

Problem: A network consists of 13 nodes A, B, C, D,

E, F, G, H, I, J, K, L and M (See Figure 3) and each

of them have 10, 20, 15, 5, 20, 10, 15, 20, 25, 15, 6,

15 and 25 amount of resources, respectively. Each of

these nodes requires 25, 15, 5, 4, 30, 8, 5, 30, 15, 10,

20, 10 and 15 amount of resources.

We demonstrate EI-technique functioning to solve

the above problem by executing following 6-phases:

(P1) Analyze the problem, make clear problem state-

ment and identify all inputs and outputs:

(i) Analyze the problem: (1) There are 13 nodes

in the network A to M, (2) Resources available at

nodes A, B, C, D, E, F, G, H, I, J, K, L and M are

Figure 3: Scenario of a network with 13 nodes.

10, 20, 15, 5, 20, 10, 15, 20, 25, 15, 6, 15 and 25,

respectively, (3) Required resources at each node

are: 25, 15, 5, 4, 30, 8, 5, 30, 15, 10, 20, 10 and

15.

(ii) Problem statement: (1) Choose a node (say A)

to run the EI technique (2) Create clusters with the

nodes having high degree of connectivity. (3) Col-

lect excess resources from its cluster nodes and

from cluster head nodes A, E, H and K. (4) Check

the genuinity of resource requirements using his-

tory. (5) Allocate required resources at nodes A,

E, H and K.

(P2) Build an EI-network for the given problem by

deploying agents at each cluster head by EIT:

Figure 4: Clusters formation of resource allocation prob-

lem.

• Cluster formation (See Figure 4): Algorithm 1

is used to create the following 4 clusters of the

given problem network. EIT forms cluster 1 at

node A, cluster 2 at node H, cluster 3 at node E

and cluster 4 at node K as shown in Figure 4.

• EI-network formation: The EIT running on

node A uses Algorithm 2 to create an EI-

network and deploys agents for interaction to

share the analyzed information of the given

problem (given by Phase 1) as shown in Figure

Figure 5: EI-network of resource allocation problem.

(P3) Estimate practical solution of a problem

(i) Resource collection for estimating solution of

a problem: (1) EIT running on node A indicates

its resource requirements parallelly to cluster head

Emergent Intelligence: A Novel Computational Intelligence Technique to Solve Problems

Table 1: Comparisons of EI, MAS and SI.

Multiagent System Emergent Intelligence Technique Swarm Intelligence

Presence of decentralization control Presence of partial decentralization and cen-

tralization control

Presence of decentralization control

Self-aware Group-aware Self-aware

Provides local view provides global view Provides local view

Takes individual decisions Takes collaborative dynamic decisions Takes local indepedent decisions

Individual interactions Group Interactions One to one interactions

Concerns management of agents behavior, like

migration, interaction, mobility, etc.

Concerns management of agents behavior, like

migration, interaction, mobility, etc. and man-

agement of information, such as resources, ac-

tivities, etc.

Concerns management of independent agents

behavior, like migration, interaction, mobility,

etc.

nodes E, H and K, (2) Nodes E, H and K paral-

lelly estimate their excess resources and they pro-

vide to node A, (3) Each agent refers to history

database parallelly to estimate accurate require-

ments of each node, and (4) Each cluster head

node parallelly shares its excess resources to the

requested cluster head nodes depending upon the

history database.

(ii) Practical solution at node A refers to history

database which gives an accurate solution and is

given as

i,A

requir

i,C

+ max(H

i,C

(t) − R

requir

i,C

, 0)

avail

i,C

+ δR

exc

i,C

(2)

where (a) δ is scaling factor and its value is 1

when R

requir

> R

avail

and is 0 otherwise, (b) R

requir

i,C

is the i

resource requirements by node A in clus-

ter 1, and (c) R

exc

i,C

(t) is the i

excess re-

sources available in cluster 1 (C

), 2 (C

), 3 (C

)

and 4 (C

) is given as

exc

i,C

(t) =

∑

c=1

∑

j=1

∆(R

avail

c, j,i

(t) − [R

pred

c, j,i

(t) + R

alloc

c, j,i

(t)])

where ∆ = 1 when R

requir

< R

avail

and ∆ = 0 oth-

erwise whereC is the total number of cluster in the

given network, n is the total number of resources

and H

i,C

(t) is the i

resource history database

of node A in cluster 1.

(P4) Create an action plan

(i) Cluster head node A parallelly indicates re-

source requirements to all other cluster head

nodes in the EI-network, (ii) Cluster head nodes

(other than A) check genuinity of resource re-

quirements by referring to the history database,

(iii) Node A negotiates resource requirements

with nodes E, H and K, and collects negotiated

excess resources, and (iv) Node A allocates re-

sources using available and collected resources

and is given as

alloc

i,A

(t) = n(P) ∗

avail

i,A

(t) +

∑

c=1

∑

i=1

coll

i,c

(t)

(1+n) ∗ H

c,i,A

(t)

(3)

where n(P) is the predicted demands at node A.

(P5) Implementation of an action plan:

(i) EIT makes cluster head nodes (A, E, H and

K) agent to cooperate, coordinate, collaborate

and negotiate among them to collect required re-

sources and support for achieving their goals, and

(ii) Resources are allocated at cluster head node

(P6) Whenever a change in input repeat phases from

1-5: In this phase, if any changes in inputs like oc-

currence of new events and resources along with

the change in external environment condition it

makes to repeat phases from 1-5 again.

5 EXAMPLE 2: JOB SHOP

SCHEDULING PROBLEM

Problem: A network consists of 15-nodes A, B, C,

D, E, F, G, H, I, J, K, L, M, N and P (See Figure

6), resources available in these nodes are 10, 20, 15,

5, 20, 10, 15, 10, 5, 10, 15, 10, 15, 10 and 15, re-

spectively. Jobs available at nodes A, D and E are

, J

and J

, respectively and these jobs need to be

schedule in a speciﬁc order. Each of these jobs require

a speciﬁc amount of resources, that is J

requires 30,

requires 20 and J

requires 40, before scheduling at

their respective nodes. We demonstrate EI-technique

Figure 6: Scenario of a network with 15 nodes.

functioning by solving the above problem by execut-

ing following 6-phases:

(P1) Analyze the problem, make clear problem state-

ment and identify all inputs and outputs

(i) Analyze the problem: (1) There are 15 nodes

in the network A to I, (2) Resources available at

ICAART 2019 - 11th International Conference on Agents and Artiﬁcial Intelligence

nodes A, B, C, D, E, F, G, H, I, J, K, L, M, N and

P are 10, 20, 15, 5, 20, 10, 15, 10,5, 10, 15, 10,

15, 10 and 15, respectively, (3) There are 3 jobs

, J

and J

at nodes A, D and E, respectively,

and (4) Job J

at node A requires 30, J

at node D

requires 20 and J

at node E requires 40 amount of

resources for scheduling at their respective nodes.

(ii) Problem statement: (1) Choose a node (say

A) to run the EI technique, (2) Create clusters

with the nodes having high degree of connectivity,

(3) Collect excess resources from its cluster nodes

and from cluster head nodes A, E, D, N and K, (4)

Check the genuinity of resource requirements us-

ing history, and (5) Allocate required resources at

nodes A, E, D, N and K.

(P2) Build an EI-network for the problem by deploy-

ing agents at each cluster head using EIT:

(i) Cluster formation: The cluster formation Al-

Figure 7: Clusters formation of job scheduling problem.

gorithm 1 is used to create following 5 clusters of

the given problem network. EIT forms cluster 1

at node A, cluster 2 at node D, cluster 3 at node

E, cluster 4 at node N and cluster 5 at node K as

shown in Figure 7.

(ii) EI-network formation: EIT running on node

A uses Algorithm 2 to create the EI-network and

deploys agents for interaction to share analyzed

information of the given problem (given by Phase

1) as shown in Figure 8.

Figure 8: EI-network of the job scheduling problem.

(P3) Estimate practical solution of given problem:

(i) Resource collection for estimation of solution

of the problem: (1) EIT running on node A paral-

lelly indicates its resource requirements to cluster

head nodes E, D, N and K, (2) Nodes E, D, N

and K parallelly estimates their excess resources

and they provide to node A, (3) Each agent par-

allelly refers to the history database to estimate

accurate requirement of each node, and (4) Each

cluster head node parallelly shares their excess re-

sources to the requested cluster head nodes de-

pending upon history database. (ii) Practical so-

lution of problem of job J

by referring to the his-

tory database at node A is given as

i,J

requir

i,J

+ +max(H

i, j,A

(t) − R

requir

i,J

, 0)

δR

exc

i,J

+ R

avail

i,J

(4)

where (a) δ = 1 when R

req

> R

avail

and δ = 0

otherwise, (b) R

requir

i,J

is i

resource requirement

by node A for scheduling the job 1, (c) R

exc

(t)

is excess resources available in cluster 1 (C

) is

given as

exc

(t) =

∑

j=1

∆(R

avail

j,i

(t) − [R

pred

j,i

(t) + R

alloc

j,i

(t)])

where ∆ = 1 when R

requir

< R

avail

and ∆ = 0 oth-

erwise. (d) H

i,J

(t) is the history database of

node A. (iii) Similarly for the jobs J

and J

nodes D and E, respectively are also estimated.

(P4) Create an action plan:

(i) Cluster head nodes D, E, K and N coordinates

with each other, and provides the maximum re-

sources to node A, (ii) These cluster head nodes

check genuinity of resource requirements by re-

ferring to the history database of node A, (iii)

Node A negotiates resource requirements with

nodes D, E, K and N, and collects negotiated

excess resources, (iv) Collected excess resources

and available resources at node A are used for re-

source allocation to schedule the job J

, and is

given as

alloc

i,A

(t) = n(T

) ∗

avail

i,A

∑

m=1

coll

i,C

(t)

(1+n) ∗ H

i,A

(t)

(5)

where n(T

) is the number of sub-tasks of job J

at node A. (v) For scheduling of the Job J

, the

resources need to be allocated to it is given as

alloc

i,D

(t) = n(T

) ∗

avail

i,D

∑

m=1

coll

i,C

(t)

(1+n) ∗ H

i,D

(t)

(6)

where n(T

) is the number of sub-tasks of job J

at node D. (vi) For scheduling of the Job J

, the

resources need to be allocated to it is given as

alloc

i,E

(t) = n(T

) ∗

avail

i,E

∑

m=1

coll

i,C

(t)

(1+n) ∗ H

i,E

(t)

(7)

where n(T

) is the number of sub-tasks of job J

at node E.

(P5) Implementation of action plan:

EIT makes cluster head nodes (A, E, D, N and

K) agent to cooperate, coordinate, collaborate

and negotiate among them to collect required re-

sources and support for achieving their goals.

• Scheduling of Job J

at node A: Cluster head

node A parallelly collects excess resources of

nodes in its cluster, allocates required resources

and schedules the job J

to the desired destina-

tion.

Emergent Intelligence: A Novel Computational Intelligence Technique to Solve Problems

• Scheduling of Job J

at node D: Cluster head

node D parallelly collects excess resources of

nodes in its cluster, allocates required resources

and schedules the job J

• Scheduling of Job J

at node E: Cluster head

E parallelly interacts with all nodes of cluster

3, collects their excess resources, allocates re-

quired resources and schedules the job J

(P6) Whenever changes in input value repeat phases

from 1-5:

In this phase, occurrences of any new jobs and

resources along with changes in external envi-

ronment condition makes to repeat phases from

1-5 again.

6 SIMULATION AND RESULTS

ANALYSIS

In this section, we describe the scenarios (Figures 3

and 6) and performance assessment of the EIT and

MAS by means of conducting the simulation experi-

ments by integrating NS-2 with MobileC agent plat-

form (B. Chen, et. al., 2010, mobilec, 2015). In the

subsequent subsections, we describe simulation sce-

narios, results analysis and comparisons.

6.1 Simulation Scenario

We have considered 13 and 15 number of nodes sim-

ulation scenario in NS2. We have measured the per-

formance of EIT and MAS by running the simulation

600 seconds. The MobileC agent platform is installed

at each cluster head node in the given network, which

deploys static agent. The static agent creates and dis-

patches mobile agents to each cluster head node in

the given problem network. In NS-2, at every nodes

resources are stored in a trace ﬁle and given to Mo-

bileC agent platform, where agents collect resources

and provides to the neighborhood cluster head nodes.

In NS-2 we have created cluster (which consists

of all neighborhood nodes), in each cluster all nodes

interact with each other. Agents are deployed by us-

ing MobileC agent platform, which makes to collect

and share available, analyzed, collected and histori-

cal resource information. The performanceevaluation

of the proposed system carried out using the simula-

tion parameters as shown Table 2 on a dual-CPU Intel

Core i5-2400 at 3.10 GHz Desktop computer with 12-

GB RAM running Fedora version 25.

Table 2: Simulation Parameters.

Parameter Values

Simulation area 2400 × 2400

Simulation time 600 sec.

Network simulation NS-2

Agent platform Mobile-C

Number of mobile agents 50

Number of static agents 9

Number of nodes) 15

Communication technology IEEE 802.11a

Network transfer rate 2 Mbps

Data processing rate 100 Mbps

6.2 Results Analysis

In this subsection, we discuss performance analysis of

MAS and EIT in job scheduling and resource alloca-

tion problems.

Throughput is the amount of information received

successfully at the cluster head node from the agent,

which is measured in bits per second (bps). In the

simulation, the amount of information received suc-

cessfully and lost during migration of agents from

one node to other to reach the desired cluster head

nodes are determined and generated the data values at

varied time intervals is given in Table 3. Simulation

result 9 shows the variation of throughput of MAS,

SI and EIT at varying simulation time. Throughput

shows the signiﬁcant improvement in the case of EIT

as compared to MAS and SI. Figure 10(a) shows data

100 150 200 250 300 350 400 450 500 550 600

Simulation Time (in sec.)

Throughput (Mbps)

MAS

Figure 9: Throughput.

collection time needed by MAS and EIT. When an in-

dividual agent is present than both technique will take

same amount of time to collect data. When there will

be set of agents, then the MAS takes more time for

data collection, because in the beginning these agents

considers only their own goal and later they collab-

orate among them. But in the case of EIT, group of

neighbor agents from the beginning cooperates and

collaborates to achieve the system goal rather than the

individual goal. Hence, EIT takes less time compared

to MAS for data collection.

Due to the diverging interest of each agent in

MAS, stores all resources till completion of job and

hence it reserves space. Therefore, the space com-

plexity of MAS is O(n

+ m), where n is total num-

ber of agents and m is total number of resources used

for execution of problem tasks t. EI technique agents

have converging interests and they release resources

after completion of each task and hence requires less

ICAART 2019 - 11th International Conference on Agents and Artiﬁcial Intelligence

100

space compare to MAS. Therefore, the space com-

plexity of EI becomes O(n + log(m)). Figure 10(b)

shows the memory consumption by each agent during

problem solving process in the MAS and EIT. Due

to the divergence and resource reservation policy in

the MAS consumes more memory than the EIT and is

shown in Figure 10(b). Figure 10(c) shows the time

complexity for taking dynamic decisions by MAS and

EIT. In MAS, each agent takes their own decision for

achieving their self-goal, therefore for achieving the

complete system goal takes O(n

+ m + t) time and

EIT based decision takes O(n

+ log(m + t)). Be-

cause in the case of EIT all agents collaborate and

cooperate depending upon requirements and abilities

for achieving whole system goal. Figure 10(d) shows

resource allocating using MAS and EIT. It shows

clearly that amount of resources available, required,

estimated accurate requirements and allocation for 3-

different tasks. Figure 10(e) shows execution time of

3-tasks of resource allocation problem. MASs have

either diverging information or interests, or both and

reserves the resources till completion of resource al-

location to a task. In EIT, execution time is less as

compared to MAS because tasks are self-organized

in such away that there should not be any conﬂicts

for resource allocation during execution of tasks in-

dependently. Self-organization realizes global goal

of system during random ﬂuctuations of environment

or conditions. Figure 10(f) shows time taken to self-

organize cluster of agents to achieve global goal of

system during any change in environment.

MAS will adapt using current available data and

history information for executing set of tasks. EIT

makes agent to adapt continuously not only using his-

tory database and current available data, and also us-

ing neighbor agents current and history information

for executing set of tasks. Figure 10(g) sows that

MAS takes less adaptability time than EIT.

We have considered 3-jobs and estimated the ex-

ecution time require to arrange and schedule them in

proper order to their desired destination and is shown

in Figure 10(h). MAS takes more execution time

for job scheduling problem because it re-executes the

same job twice when we decompose its problem into

sub-problems and execute them independently. In

the case of EIT, it decomposes the jobs problem in

such away that they should not be any dependency on

others. Hence, EIT parallelly executes decomposed

problems of jobs and it requires less execution time

as compare to MAS.

Table 3: Throughput of EIT, MAS and SI.

Simulation Time (in

sec.)

MAS SI EIT

100 9.4367 9.5846 10.6734

200 10.0134 10.2643 10.80143

300 11.6038 11.2365 12.49871

400 12.1764 12.3251 14.4768

500 11.4379 11.6982 13.0179

600 11.3127 11.4523 12.8263

7 CONCLUSIONS

We presented the novel EIT for solving problems in

uncertain environments and explained the function-

ing and phases involved. The resource allocation

and job-shop scheduling problems have solved us-

ing EIT. Analyzed the performances of EIT, MAS

and SI by considering performance measures, such

as space complexity, time complexity, data collection

time, throughput, time taken for self-organization and

adaptation. These performance measures are shown

that the EIT performance better than the MAS and SI

for solving problems in uncertain environments.

REFERENCES

Jacak, Witold, and Karin Pr¨oll. ”Heuristic approach to con-

ﬂict problem solving in an intelligent multiagent sys-

tem.” In International Conference on Computer Aided

Systems Theory, pp. 772-779. Springer, Berlin, Hei-

delberg, 2007.

Gundersen, Odd Erik, and Anders Kofod-Petersen. ”Multia-

gent based problem-solving in a mobile environment.”

Norsk Informatikkonferance (2005): 7-18.

Maryam Rajabi, Teh Noranis Mohd Aris, “The Role

of Multi-Agent in Computational Problem Solving

Environments”, In the 4th International Conference

on Electrical Engineering and Informatics, Procedia

Technology, 2013, pp.1103-1109.

Sun, S., Joy, M. & Grifﬁths, N. The Use of Learning Ob-

jects and Learning Styles in a Multi-Agent Education

System. Journal of Interactive Learning Research; 18

(3); 2007. p. 381.

Fenza, G., Furno, D., Loia, V. Hybrid Approach for

Context-aware Service Discovery in Health care Do-

main. Journal of Computer and System Sciences 78;

2012. p. 123.

Pakonen, A., Tommila, T., Pirttioja, T., Selionen, I. OWL

based information agent services for process monitor-

ing. Emerging Technologies and Factory Automation

( ETFA, IEEE Conference) , Patras, Greece; 2007. p.

9–16.

Bell, D. A., Beck, A., Miller, P., Wu, Q. X., Herrera. A.:

Video Mining –Learning Patterns of Behavior via an

Intelligent Image Analysis System. The 7th Interna-

tional Conference on Intelligent Systems Design and

Applications; 2007. p. 460-464.

Emergent Intelligence: A Novel Computational Intelligence Technique to Solve Problems

101

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Data collection (in %)

Data collection time (in minutes)

Individual agent (both MAS and EI)

Set of agents (MAS)

Set of agents (EI)

(a) Data collection time needed

by MAS and EI

1 2 3 4 5

100

120

140

160

180

200

Number of agents

Memory consumption (in bytes)

MAS

(b) Memory consumption

by each agent during solving

problem: MAS and EIT

0 1 2 3 4 5 6 7 8 9

Time (in sec.)

Time required to complete a decision (in sec.)

Emergent Intelligence

MAS

lem: MAS and EIT

1 2 3

Tasks

Amount of resources

MAS based resource allocation

EI based resource allocationn

Required resources

Available resources

Accurate resource requirement

(d) Resource allocation using MAS

and EI technique

1 2 3

Number of tasks

Execution time (in seconds)

EI based task execution

MAS based task execution

(e) Taks execution time using MAS

and EIT:Resource allocation problem

0 5 10 15 20 25 30 35 40 45 50

100

120

140

160

180

200

220

240

260

Number of agents

Time taken for self−organization(in seconds)

EI based self−organization

(f) Time taken for self-

organization of clusters of a

problem

0 5 10 15 20 25 30 35 40 45 50

100

120

140

160

180

200

220

240

260

Number of agents

Time taken for adaptability(in seconds)

EI based adatability

MAS based adptability

(g) Adaptability of MAS and EIT

for problem solving

1 2 3

Number of jobs

Execution time (in seconds)

MAS based jobs execution

EI based jobs execution

(h) Execution time using EI and MAS:

Job scheduling problem

Figure 10: Simulation and Analysis Results.

De Weerdt, M., Bos, A., Tonino, H., and Witteveen, C. A

resource logic for multi-agent plan merging. Annals

of Mathematics and Artiﬁcial Intelligence 37 ( 1–2);

2003. p. 93.

Bala M. Developing Intelligent Agent Applications with

JADE and JESS. Springer-Verlag Berlin Heidelberg

Knowledge-Based Intelligent Information and Engi-

neering Systems Lecture Notes in Computer Science.

5179; 2008. p. 236-244 .

Kordaki, M. A drawing and multi-representational com-

puter environment for beginners’ learning of program-

ming using C: Design and pilotformative evaluation.

Computers & Education 54(1); 2010. p. 69-87.

Aris, T.N. An Agent-Based Conceptual Model For Compu-

tational Problem Solving. The International Confer-

ence on Informatics and Applications (ICIA2012) –

Malaysia; 2012. p. 240-248.

Reynolds, Joshua, Lauren Christopher, Russ Eberhart, and

Patrick Shaffer. ”Using computational swarm intel-

ligence for real-time asset allocation.” In Computa-

tional Intelligence for Security and Defense Applica-

tions (CISDA), 2015 IEEE Symposium on, pp. 1-5.

IEEE, 2015.

Johansson, F.; Falkman, G., ”Real time allocation of defen-

sive resources to rockets, artillery, and mortars,” Infor-

mation Fusion (FUSION), 2010 13th Conference on,

vol., no., pp.1-8.

Kalyan V., and Lisa Ann Osadciw. ”Optimal scheduling in

sensor networks using swarm intelligence.” In Pro-

ceedings of 38th Annual Conference on Information

Systems and Sciences, pp. 17-19. Princeton Univer-

sity Princeton, New Jersey, 2004.

Guido Guizzi1, Francesco Gargiulo1, Liberatina Carmela

Santillo1 , and Hamido Fujita, “ Swarm Intelligence

in Evacuation Problems: a Review”.

Rzevski, G. and Skobelev, P, “Emergent intelligence in

large scale multi-agent systems”. International Jour-

nal of Education and Information Technology, 2007,

Vol. 1(2), pp.64-71.

W. D. Hillis, ’Intelligence as an emergent behavior; or, the

songs of eden’, Daedalus (1988), pp. 175–189.

T. Wolf, ’Analyzing and engineering self-organizing emer-

gent applications’, 2007.

Z. Li, C. H. Sim, M. Y. H. Low, ’A survey of emergent

behavior and its impacts in agent-based systems’, in:

Industrial Informatics, IEEE International Conference

on, 2006, pp.1295-1300.

P. J. Angeline, ’Genetic programming and emergent in-

telligence’, Advances in genetic programming 1994,

Vol.1, pp.75-98.

B. Chen, H. H. Cheng, ’A review of the applications

of agent technology in trafﬁc and transportation

systems’, Intelligent Transportation Systems, IEEE

Transactions on 11 (2) (2010) pp. 485–497.

’A multi-agent platform for mobile c/c++ agents’,

http://www.mobilec.org/overview.php, accessed on

June, 2015.

ICAART 2019 - 11th International Conference on Agents and Artiﬁcial Intelligence

102