Cognitive Computing Meets the Internet of Things

Zakaria Maamar

, Thar Baker

, Noura Faci

, Emir Ugljanin

, Yacine Atif

Mohammed Al-Khafajiy

and Mohamed Sellami

Zayed University, Dubai, U.A.E.

Liverpool John Moores University, Liverpool, U.K.

Universit´e Lyon 1, Lyon, France

State University of Novi Pazar, Novi Pazar, Serbia

University of Sk¨ovde, Sk¨ovde, Sweden

ISEP Paris, Paris, France

Keywords:

Business Process, Cognitive Computing, Internet-of-Things.

Abstract:

This paper discusses the blend of cognitive computing with the Internet-of-Things that should result into

developing cognitive things. Today’s things are conﬁned into a data-supplier role, which deprives them from

being the technology of choice for smart applications development. Cognitive computing is about reasoning,

learning, explaining, acting, etc. In this paper, cognitive things’ features include functional and non-functional

restrictions along with a 3 stage operation cycle that takes into account these restrictions during reasoning,

adaptation, and learning. Some implementation details about cognitive things are included in this paper based

on a water pipe case-study.

1 INTRODUCTION

According to a 2015 IBM white-paper (Green, 2015),

Internet-of-Things (IoT) needs to be smarter so that

better results from things could be attained. This

smartness could become effective thanks to cognitive

computing. In a similar statement, Wu et al. argue

that “without comprehensive cognitive capability, IoT

is just like an awkward stegosaurus: all brawn a nd

no brains” (Wu et al., 2014). Brain-empowered IoT

or Cognitive Internet-of-Things (CIoT) are the terms

that Wu et al. use to describe the future generation of

things. In line with the cognitive trend, a 2017 ana-

log devices white-paper states that “The Internet of

Things Depends on the Intelligence of Things”

Tapping into the opportunities of IoT by, for in-

stance, offering better services through thing compo-

sition, organizations, also, rely on Business Proces-

ses (BP) to achieve their missions. A BP “...is no -

thing more than the coding of a lesson learnt in the

past, transformed into a standard by a group of ex-

perts and established as a mandatory ﬂow for those

who must effectively carry out the work” (OpenKno-

wledge, 2016).

Despite the “hype” surrounding IoT, the ICT com-

www.mouser.com/pdfdocs/Technologies-and-

Applications-for-the-IoT.pdf.

munity is somehow not “satisﬁed” with the passive

nature of things due to their current role in mainly

supplying data (DZone, 2017; Mzahm et al., 2013).

To address this nature, we examine the blend of

cognitive computing with IoT in the particular con-

text of BP. Injecting cognitive capabilities into IoT

would result into Cognitive Things (CT) that BP would

have to interact with (i.e., not act-upon things nor di-

rect things like discussed in (Haller and Magerkurth,

2017; Suri et al., 2017)) according to ﬁrst, these BPs’

business logics’ needs and requirements and second,

the context of these CT. Our objective is to empower

things with reasoning, learning, and adaptation capa-

bilities, so that, a BP would weave these things into

its process model. Though some might be skeptical

about thing empowerment, Taivalsaari and Mikkonen

argue that “hardwa re adv ances an d the availability of

powerful but inexpensive integrated chips will make

it possible to embed connec tivity and fully edged vir-

tual machines and dyna m ic language run-times ever-

ywhere” (Taivalsaari and Mikkonen, 2017). As a re-

sult of these advances, everyday things will become

connected and programmable dynamically.

Section 2 brieﬂy presents the concepts of IoT

and cognitive computing and suggests a case study.

Section 3 is how to put the blend of cognitive compu-

ting with IoT in the context of BP into action. Some

Maamar, Z., Baker, T., Faci, N., Ugljanin, E., Atif, Y., Al-Khafajiy, M. and Sellami, M.

Cognitive Computing Meets the Internet of Things.

DOI: 10.5220/0006877507410746

In Proceedings of the 13th International Conference on Software Technologies (ICSOFT 2018), pages 741-746

ISBN: 978-989-758-320-9

741

preliminary implementation results are reported in

Section 4. Finally, Section 5 concludes the paper.

2 BACKGROUND

Internet of Things. The abundant literature on IoT

does not help propose a unique deﬁnition of what

IoT is or should be. On the one hand, Barnaghi and

Sheth provide a good overview of IoT requirements

and challenges (Barnaghi and Sheth, 2016). Require-

ments include quality, latency, trust, availability, re-

liability, and continuity that should impact efﬁcient

access and use of IoT data and services. And, chal-

lenges result from today’s IoT ecosystems that fea-

ture billions of dynamic things that make existing se-

arch, discovery, and access techniques and solutions

inappropriate for IoT data and services. On the other

hand, Abdmeziem et al. discuss IoT characteristics

and enabling technologies (Abdmeziem et al., 2016).

First, characteristics include distribution, interopera-

bility, scalability, resource scarcity, and security. Se-

cond, enabling technologies include sensing, commu-

nication, and actuating. These technologies are map-

ped onto a 3 layer IoT architecture that consists of per-

ception, network, and application, respectively.

Cognitive Computing. Sheth, in (Sheth, 2016), re-

fers to DARPA’s deﬁnition of cognitive system as a

system that can “reason, use represented knowledge,

learn from experience, accumulate knowledge, ex-

plain itself, accept direction, b e aware of its own be-

havior and capab ilities as well as re spond in a robust

manner to su rprises” (Johnson, 2002). This deﬁni-

tion identiﬁes some capabilities that could empower

things such as learning and sensing. According to

Raut

, cognitive computing systems may include dif-

ferent components such as natural language proces-

sing, machine learning, image recognition, and emo-

tional intelligence.

Case Study. It is about cognitive water-pipes in sup-

port of smart homes’ services. It is well known that

leaks are a signiﬁcant source of water loss. However,

it is less known that a large proportion of this loss, 20-

30%, occurs at the consumer side. According to the

Association of British Insurers Research, the average

cost from a burst pipe is £6,500 to £7,500 (cas, ). On

top of this cost, insurance companies spend billions to

cover water damages and cost of repairs.

We, safely, assume that walls in today’s smart ho-

mes have mounted moisture detecting sensors, which

could help reduce water loss and hence, bills. The

bigdata-madesimple.com/what-exactly-is-cognitive-

computing.

sensors would alert tenants of any water pipe leakage

before it leads to serious damages. However, by the

time the tenant notices the alert, then ﬁnding a plum-

bing company to book for repair, the wall itself could

end up costing some money to get ﬁxed, for example.

Our proposal is that cognitive water-pipes would

reason about sensed data (e.g., leak position and time

it started, amount of drippings, and moisture level) so,

they, for instance, ask the water distribution company

to suspend water provisioning, contact potential re-

pair services to come ﬁx the leak, and ﬁnally, make

a payment. In this case, searching for and calling re-

pair services, negotiating deals with them, and ma-

king contact with the tenant’s bank account to com-

plete a service payment are all individual BP that rely

on CT engagement in addressing water pipes’ leaks.

3 HOW TO ACTION THE BLEND?

3.1 Features of Cognitive Things

We empower a CT with 3 types of capabilities (not ne-

cessarily all) that would allow this CT to reason about

the surrounding, to learn from the past, and to adapt

to changes. These capabilities include computation

for processing needs, persistence for storage needs

(even temporarily), and communication for transfer

needs. The enactment of each capability is subject to

2 types of restrictions on the CT: functional and non-

functional.

Functional restrictions impact a CT participation

in ongoing BP (in fact, BP instances at run-time). We

decompose these restrictions into 3 categories:

- Limited (l): when a CT participation is restricted

by a time frame. Beyond this time frame, the CT

ceases to exist (e.g., withdrawn because of expiry

date) and hence, becomes unavailable for certain

BP (however, the CT would remain available for

other BP). Example of limited is a moisture sensor

that has a life span due to power availability (on

battery) and/or part deterioration over time.

- Non-shareable (ns): when a CT concurrent par-

ticipation in many BP needs to be scheduled

(e.g., required because of conﬂicting requests).

Example of non-shareable is a water meter de-

dicated to personal usage and hence, cannot be

shared with other residential units.

- Renewable (r): when a CT participation in a BP

is extended for another time frame and/or round

of use subject to satisfying the limited and/or sh-

areability restrictions (e.g., approved because of

work incompleteness). Example of renewable is a

ICSOFT 2018 - 13th International Conference on Software Technologies

742

2 hour-rented pump to drain water. However, the

rent can be extended, if necessary.

Non-functional restrictions impact a CT participa-

tion in ongoing BP in terms of processing power,

storage capacity, and/or communication bandwidth.

- Processing (p) is about minimum versus maxi-

mum number of instructions.

- Storage (s) is about limited versus unlimited and

persistence versus volatile.

- Communication (c) is about minimum versus

maximum data transfer.

3.2 Operations over Cognitive Things

We propose an ecosystem of CT that is built-upon

3 connected worlds (Fig. 1): the process world fea-

turing BP, the thing world featuring CT, and the data

world featuring data linked to both BP and CT. As

stated in Section 1, a BP neither act upon a CT nor

direct it. Contrarily, BP and CT engage in continu-

ous interactions that should, ideally, lead to conﬁr-

ming the participation of CT in BP as well as trigge-

ring new BP. A participation considers a CT’s functio-

nal and non-functional restrictions that, in fact, reﬂect

this CT’s current/active participation in other ongoing

(under-execution) BP. Still in Fig. 1, the thing world

produces data

(e.g., after sensing) that the process

world manages in terms of consuming these data an-

d/or producing new data. Managing data would make

BP (i.e., instances) progress in their executions along

with initiating additional interactions with new and/or

(some) current CT and/or closing ongoing interactions

with (some) current CT.

Process world Thing world

Data world

produce

interact

Cognition

manage

Figure 1: Ecosystem of cognitive things.

In Fig. 2, we suggest a 3 stage cycle for deﬁ-

ning the cognition anchored to the world of things in

Fig. 1 (cloud shape). In the reasoning stage, a CT

assesses the surroundings (e.g., context) on top of its

functional and non-functional restrictions prior to ma-

king any new decision of participating in another BP

Data issues like semantics do not fall into the scope of

this work.

(the BP also considers its ﬁnancial restriction) or con-

tinuing (in compliance with the renewal functional-

restriction) its participation in an ongoing BP. To this

end, the CT relies on both the data in the data world

and the respective statuses of all ongoing interactions

with the process world. Some decisions in the rea-

soning stage could lead to conﬁrming CTs’ participa-

tion in BP and adjusting CTs’ behaviors (e.g., can-

celing a participation in a BP) as per the adapta-

tion stage (i.e., changes in behaviors (Terdjimi et al.,

2017)). Lessons learned during the adaptation stage

feed the learning stage that itself feeds the reasoning

stage with details on these lessons. Examples of de-

tails could be the number of times a CT participation

in a BP has been renewed (in compliance with the re-

newable functional-restriction).

Reasoning

Adaptation

could lead to

could help

could feed

Learning

Cognitive

thing

Figure 2: Cognition of IoT as a 3 stage cycle.

Let us apply the above 3 stage cycle to the water-

pipe case study (Fig. 3). First, when a leak is detected,

the moisture sensor CT generates data like amount of

drippings and moisture level and decides (reasoning,

Section 3.3) about the severity of leak and hence, the

urgency of ﬁxing the pipe. If it is not severe, the

sensor CT informs the tenant of the leak. Contrarily,

the sensor CT triggers a new pipe ﬁxing BP. This BP

requires checking if the maintenance contract CT is

still valid (in compliance with the limited functional-

restriction) as it can be extended, if necessary (rene-

wal taken care by adaptation). The contract CT men-

tions an agreed-upon plumbing company that will do

the necessary job. In conjunction with contacting the

plumbing company, the moisture CT informs the me-

ter CT to close the water distribution due to past cases

that led to neighbors’ complaints (reasoning). Feed-

back on the quality of repair permits to update the

maintenance contract CT (learning).

3.3 Reasoning of Cognitive Things

Since CT are resource-bounded, we adopt the Belief-

Desire-Intention (BDI) approach (Bratman, 1987) to

represent a CT’s cognition. CT are empowered with

reasoning capabilities that tap into recurring events

Cognitive Computing Meets the Internet of Things

743

Moisture CT

Maintenance

contract CT

Meter CT

1. leak

detection

Business Process

3. initiation

4. contact

(not-severe)

5. distribution closure (severe)

2. leak

severity

7. contract

validity

6. initiation repair

(severe)

8. repair initiation (valid)

Tenant

Plumber

9. contract

(invalid)

Wall

Figure 3: Illustration of the 3 stage cycle.

and subsequent course of action

to assert certain

beliefs (knowledge about the CT’s context). Be-

liefs can be represented as a collection of properties

(i.e., beliefs, goals/desires, and intentions/plans) cap-

tured through predicate-logic statements (i.e., predi-

cates and rules), formatted into some speciﬁc standard

for data exchange (e.g., JSON and XML), and sto-

red in some knowledge base. Fig. 4 depicts a thing’s

cognition as a set of beliefs and reasoning capabi-

lities (i.e., goal matching, belief revision, delibera-

tion, and plan selec tion). Beliefs are updated from

events generated by the thing world and from inte-

ractions with other CT through the belief revision ca-

pability. The belief knowledge-base can be maintai-

ned through two cognitive processes: perception and

inﬂuenc e bias. Perception refers to some transfer of

information from the process and/or thing worlds into

beliefs while inﬂuence bias refers to belief revision

based on interactions with other CT. Since beliefs are

uncertain, inﬂuenc e bia s depends on to what extent

other CT are trusted. Belief revision enables a CT to

continuously learn by curating its beliefs and updating

its decision rules. Goals can be represented as target

states that refer to some beliefs. Matching goals with

conclusion part of decision rules enables pro-active

behaviors of CT on their own. Once goal matching is

performed, a CT’s delib eration infers alternative in-

tentions by selecting appropriate plans for execution.

These plans are applied on the thing world so that go-

als are achieved. Plans are not just a sequence of ba-

sic actions, but may also generate new sub-goals.

As stated earlier, learning happens through in-

cremental belief-amendment from perception and be-

lief revision triggered by events occurred in the thing

world. This world includes devices (e.g., sensors) an-

chored to physical phenomena and linked to BP that

Note that a course of action result from an intentional

reasoning that drives a CT’s behavior.

Revise

beliefs

Update

Beliefs

Engage

in goals

Match

Deliberate

Reason

Update

Intentions

Submit

Select

plans

Act

upon

Verify

Interaction

with other CT

Thing

world

Events

Activity

Knowledge

base

Legend

Figure 4: Thing’s cognition in action

collect and curate data. The reasoning starts with

a goal engagement that satisﬁes some BPs’ require-

ments. This goal is enabled based on satisfying some

contextual conditions that feature the proposed cogni-

tion cycle described in Fig. 4. Examples of conditions

could level of humidity in a home.

In the following, we show that the knowledge of

a CT could evolve over time thanks to learning. This

evolution develops a new Belief B

′

in the form of pre-

dicates and/or decision rules. This augmented kno-

wledge base is the consequence of new information

from other CT and/or devices driven by Event E

that an existing Belief B is changed. The process of

maximizing a CT’s cognition to meet, is formally ex-

pressed in Equation 1

argmax

P(B → B

′

) (1)

In addition, Equation 2 shapes the incremental

scope of CT’s cognition, whereby the expectation of

a CT’s cognitive belief B

′

is asserted under stimulus E

exceeds the probability that CT’s belief B

′

is asserted

independently:

P(B → B

′

) > P(B → B

′

) (2)

4 PROOF-OF-CONCEPT

Our under-development CT testbed consists of the

following components: a temperature-humidity sen-

sor AM2302 (DHT22), Arduino UNO, and Rasp-

berry Pi2 Model B. For assembly needs, we procee-

ded as follows: the AM2302 sensor reads and sends

air humidity via analog signal. However, since Ras-

pberry Pi2 cannot read analog signal, we connected

the sensor to Arduino UNO so that this latter provi-

des data to Raspberry Pi2 through serial communica-

tion (i.e., over a USB connection). The testbed en-

vironment is presented in Fig. 5 and is referred to as

CT node. Raspberry Pi2 is connected to the Internet

ICSOFT 2018 - 13th International Conference on Software Technologies

744

Figure 5: CT testbed environment.

via a LAN to provide the outside world communica-

tion for the necessary BP.

From a functional perspective (Fig. 3), the test-

bed is developed to support smart-home services. It

monitors air humidity level in order to “tell” if there

is a leakage in water pipes where the CT node is in-

stalled. Therefore the developed testbed functions as

follows (Alg. 1):

1. The CT node measures the air-humidity (Hu),

Alg. 1:line 1, level every 2 minutes (line 8) (this

can be changed depending on the scenario or sy-

stem requirement). The CT node examines the

humidity level in order to check the water-pipe’s

leak status, hence we assumed the range of nor-

mal and abnormal humidity level. A humidity be-

tween 70% and 120% is treated as abnormal hu-

midity level requiring repair.

2. When a leak is detected (lines 3 & 4), the CT node

ﬁrst, returns the location of the house (based on

the latitude and longitude), and the corresponding

category of the air-humidity (line 13). The cate-

gory is used to specify the emergency level of the

required plumbing service. Then, the CT node se-

arches for the best available plumbing services in

the neighborhood. We assume that these services

are already available online so that the CT node

carries out the necessary searches based on loca-

tion, price, and tenant balance criteria (line 15,

16 & 17, respectively), for example. Then, the

CT node fetches data for these services from a

Web page.

3. Once the CT node selects the best service, it sends

out an email to the tenant so he is informed of the

issue and best available service (line 19).

4. Before the service is booked, the CT node trig-

gers a BP that compares the tenant’s account ba-

lance with the returned plumbing services best

price (line 18). If the maintenance cost can-

not be covered, the tenant is informed again by

email (line 21).

5. The CT node performs the pay (CatN), which im-

plies that the plumber has turned out and ﬁxed the

leak (line 24).

6. Finally, the CT node keeps monitoring the humi-

dity level in the pipe for 3 days (this can be altered

based on the system needs) to ensure the quality

of repair (line 25). If the category of Hu is abnor-

mal (line 26), a new appointment with the same

plumber will be arranged (line 27).

5 CONCLUSION

In this paper, we discussed the blend of cognitive

computing with the Internet-of-Things in order to fos-

ter thing seamless integration into the business world.

This blend results into cognitive things (CT) that

should be empowered with reasoning, adaptation, and

learning capabilities. These capabilities allow CT

to be active (i.e., reason, learn, and adapt) in an

ecosystem of IoT. To enable cognitive capabilities, we

ﬁrst, bind them with functional and non-functional re-

strictions along with price strategies for competition

purposes. We, also, deﬁne a 3 stage cycle governing

CT’s enactment that revolves around a BDI architec-

ture. Our under-development CT testbed consists of

a temperature-humidity sensor AM2302 (DHT22), Ar-

duino UNO, and Raspberry Pi2 Model B and has been

used in the context of leak detection in water pipes. In

term of future work, we would like to analyze on-the-

ﬂy code injection into things in compliance with the

learning and adaptation stages. Indeed, things could

be exposed to unseen situations that require new cour-

ses of action.

Cognitive Computing Meets the Internet of Things

745

deﬁnitions : Humidity (Hu), Location

(Loc), Category (Cat), eMail

(eM);

assumptions : Hu =











CatN : ∀Hu < 70

CatA : ∀70 < Hu < 80

CatB : ∀80 < Hu < 90

CatC : ∀90 < Hu < 100

CatD : ∀100 < Hu < 110

CatE : ∀110 < Hu < 120

initializ ati on : Hu=φ, Loc=φ, Cat=φ;

Get

sensorData(Hu) while Hu 6= φ do

2 Find corresponding Cat to Hu as per

assumptions;

3 if (Hu =

CatAkCatBkCatCkCatDkCatE) then

4 goto 13;

5 else

6 if (Hu = CatN) then

7 sleep (120

sec

);

8 ⊲ checks every 2min

9 goto 1;

10 end

11 end

12 end

Get

Loc(latitude, longitude, Cat) ⊲ incident

location while Loc 6= φ do

Get

localServices;

Get

bestPrice;

Get

tenantBalence;

17 if (bestPrice < tanentBalence) then

18 booking (appointment);

19 eM (tenant, booking);

20 else

21 eM (tenant, No enough credit)

22 end

23 if (CT ← appointment) && (Hu=CatN)

then

24 pay(serviceProvider, prices);

25 monitor(Cat, period 3 days);

26 if (cat 6= CatN) then

27 eM(serviceProvider,

newAppointment)

28 end

29 end

30 end

Algorithm 1: CT node process.

REFERENCES

Smart Homes: an Emerging Trend. https://

www.rsabroker.com/news/smart-homes-emerging-

trend, visited October 2017.

Abdmeziem, M., Tandjaoui, D., and Romdhani, I. (2016).

In Koubaa, A. and Shakshuki, E., editors, Robots and

Sensor Clouds, chapter Architecting the Internet of

Things: State of the Art. Springer International Pu-

blishing.

Barnaghi, P. and Sheth, A. (2016). On Searching the Inter-

net of Things: Requirements and Challenges. IEEE

Intelligent Systems, 31(6).

Bratman, M. (1987). Intention, Plans, and Practical Rea-

son. Harvard University Press, Cambridge, MA.

DZone (https://dzone.com/guides/iot-applications-

protocols-and-best-practices, 2017 (visited in

May 2017)). The Internet of Things, Application,

Protocls, and Best Practices. Technical report.

Green, H. (December 2015). The Internet of Things in the

Cognitive Era: Realizing the Future and Full Potential

of Connected Devices. www-01.ibm.com/common/

ssi/cgi-bin/ssialias?htmlﬁd=WWW12366USEN.

Haller, S. and Magerkurth, C. (xxxx (checked

out in October 2017)). The Real-time En-

terprise: IoT-enabled Business Processes.

Technical report, www.iab.org/wp-content/IAB-

uploads/2011/03/Haller.pdf.

Johnson, R. (EE Times, December 2002). Darpa

Puts Thought into Cognitive Computings.

www.eetimes.com.

Mzahm, A., Ahmad, M., and Tang, A. (2013). Agents

of Things (AoT): An intelligent operational concept

of the Internet of Things (IoT). In Proceedings of

the 13th International Conference on Intellient Sys-

tems Design and Applications (ISDA’2013), Bangi,

Malaysia.

OpenKnowledge (2012 (checked out in May 2016)).

Social Business Process Reengineering. Techni-

cal report, http://socialbusinessmanifesto.com/social-

business-process-reengineering.

Sheth, A. (March/April 2016). Internet of Things to Smart

IoT Through Semantic, Cognitive, and Perceptual

Computing. IEEE Intelligent Systems, 31(2).

Suri, K., Gaaloul, W., Cuccuru, A., and Gerard, S. (2017).

Semantic Framework for Internet of Things-Aware

Business Process Development. In Proceedings of

the 26th IEEE International Conference on Enabling

Technologies: Infrastructure for Collaborative Enter-

prises (WETICE’2017), Poznan, Poland.

Taivalsaari, A. and Mikkonen, T. (2017). A Roadmap to the

Programmable World: Software Challenges in the IoT

Era. IE EE Software, 34(1).

Terdjimi, M., M´edini, L., Mrissa, M., and Maleshkova,

M. (2017). Multi-purpose Adaptation in the Web of

Things. In P roceedings of the 10th International and

Interdisciplinary Conference on Modeling and Using

Context (CON TEXT’2017), Paris, France.

Wu, Q., Ding, G., Xu, Y., Feg, S., Du, Z., Wang, J., and

Long, K. (April 2014). Cognitive Internet of Things:

A New Paradigm Beyond Connection. IEEE Internet

of Things Journal, 1(2).

ICSOFT 2018 - 13th International Conference on Software Technologies

746