I2oT：Advanced Direction of the Internet of Things

Yixin Zhong

（University of Posts and Telecommunications，Beijing 1000876，China）

I2oT：Advanced Direction of the Internet of Things

Yixin Zhong

（University of Posts and Telecommunications，Beijing 1000876，China）

The Internet of Things（IoT）is still in its infancy because of the limited capability of its embedded processor.In the meantime，re?search on artificial intelligence（AI）has made plenty of progress.The application of AI to IoT will significantly increase the capa?bilities of IoT，and this will benefit both economic and social development.In this paper，the elementary concepts and key tech?nologies of AI are explained，and the model and principle of intelligent IoT，denoted I2oT，resulting from the integration of AI and IoT are discussed.I2oT will be the most promising version of IoT.Finally，recommendations for further study and standardization of I2oT are made.

Internet of Things；artificial intelligence；knowledge producing；strategy formulation；intelligent internet of things

1 Introduction

T here are two main motivations for expanding the In?ternet to the Internet of Things（IoT）.The first moti?vation is to expand the amount of information shared by databases and objects in the real world. The second motivation is to enable users not only to share in?formation but also control objects in the real world.These make IoT much more attractive in society.In other words，IoT is a good advancement of the conventional Internet.

In terms of technological development，however，IoT is still in its infancy and can be greatly improved by endowing IoT functions with much more intelligence［1］.Significant progress has been made in artificial intelligence（AI）over the past de?cade.All AI technologies needed to make IoT more intelligent and evolve into I2oT are now feasible.The main concern at the moment is how to understand and effectively apply AI technol?ogies to current IoT systems.

2 A Brief Description of IoT

The purpose of IoT is to expand the functions of existing In?ternet and make it more useful.With IoT，users can share not only information provided by humans and contained in databas?es but also information provided by things in physical world. The simplified functional model of IoT is shown in Fig.1.

As in Fig.1，IoT has sensors，for acquiring information about the state of things；an embedded processor，for produc?ing orders that regulate the state of things；wireless technology， for transferring information from sensors to Internet and Inter?net to controller；and control unit，for executing human orders regulating the state of things.

Take IoT for maintaining room temperature for example.A standard room temperature is designated in advance，and the actual room temperature is acquired by the sensor（s）and trans?ferred via wireless to Internet.After receiving the actual room temperature，the embedded processor compares it with the des?ignated value and generates an order to regulate the room tem?perature and keep it within a certain range.This order is imme?diately sent to the control unit via the Internet and wireless unit and is executed by the actuator of the control unit.

If information about the state of the thing concerned can be acquired by sensors and controlled by actuators，and if the function performed by the embedded processor is not too com?plicated，the IoT technology is feasible.

If physical things and their environment in IoT become com?plex，the functions of the required embedded processors also become complex，and conventional technologies of the currentIoT will no longer be satisfactory.

Unfortunately，problems with complex factors are very often important to economic and social development.A typical exam?ple is air pollution over a large area.Another typical example is global warming.People want to know information about the air quality and weather conditions and control them in certain ways.Therefore，efficiently dealing with complex problems is an unavoidable responsibility of scientists.

The most promising approach to handling such complex problems is artificial intelligent.The reason for this proposal is the fact that central need for solving complex problems is the learning ability.

3 Fundamental Concepts and Principles of Artificial Intelligence

In a narrow sense，AI has traditionally implied the simula?tion of logical human thinking using computer technology. Within this framework，the fields of artificial neural networks（ANNs）［2］-［4］and sensor?motor systems（SMSs）［5］-［7］were considered extraneous，even though both fields have been con?cerned with simulating the functions of the human brain.ANN and SMS had to form a new discipline called computational in?telligence（CI）.Computational intelligence has become the oth?er approach to AI.It is more reasonable for the term AI to en?compass both AI in narrow sense and CI.In the contemporary sense，AI is now re?termed unified AI［8］-［9］.

In this paper，AI means unified AI，a general term represent?ing the theory and technology related to simulating intellectual abilities of human being，including the ability to understand and solve problems.What follows is a brief explanation of how AI can handle complex problems［10］-［12］.

What AI simulates and offers is not anything else but the learning ability of human beings，i.e.，learning to understand and solve the problem.Therefore，learning is the central fea?ture in AI and learning?technology is the key to handling prob?lems.

The simplest model for AI is roughly abstracted in Fig.2.

Ontological information（OI）in Fig.2 is information about the state and pattern of the state variance that are presented by the object in the environment of the outside world and that are the resources and clues for learning to understand the prob?lem.On the other hand，the subject’s action or reaction ap?plied to the object can be learnt based on an understanding of the problem.

A more specific functional model of the technologies in AI is shown in Fig.3.In Fig.3，AI technologies are interconnect?ed and interact with each other.

3.1 Categories of AI Technology

3.1.1 Perception

This technology is used to acquire the OI about the object or problem in its environment.It is also the technology for con?verting OI to epistemological information（EI）.

Epistemological information is information perceived by the subject about the trinity of the form（syntactic information），content/meaning（semantic information），and utility/value（pragmatic information）concerning OI.

Unlike the traditional concept of information proposed by Claude Shannon，EI comprises the trinity of the form，content/ meaning，and utility/value and is the basis of learning.This is why EI is also often called comprehensive information.

The essential function of perception is to convert OI to EI. This is the first class of information conversion in AI.

3.1.2 Cognition

The main function of cognition technology is to convert EI，which is perceived by the subject from OI，into the correspond?ing knowledge about the object.This is the second class of in?formation conversion needed in AI.The only possible ap?proach to converting EI to knowledge must be learning—there is no other way.

3.1.3 Decision?Making

The technology used in decision?making converts EI to intel?ligent strategy（IS）based on knowledge support and is directed by the goal of problem solving.The strategy is just the proce?dural guidance for problem?solving.This is the third class of information conversion in AI.

The radical function of decision?making technology is learn?ing to find the optimal solution for a given problem.There are usually a number of ways of achieving the designated goal froma starting point expressed by EI.A decision should be made through intelligent use，via learning，of the relevant knowledge provided.

3.1.4 Strategy?Execution

This technology is used to convert the IS into intelligent ac?tion（IA）that will solve the problem.

3.1.5 Strategy?Optimization

Because of various non?ideal factors in all sub?processes in Fig.3，there are often errors when intelligent action is applied. These errors are regarded as new information and are fed back to the input of the perception of the model.With this new infor?mation，the knowledge can be improved via learning，and the strategy can be optimized.Such an optimization process might continue many times until the error is sufficiently small.

In sum，all the AI technologies hereto mentioned are learn?ing?based，and this is why AI is powerful.

3.2 Implementation Issues for the Three Classes of Information Conversion

Perception technology can be implemented using the model in Fig.4，which converts OI to EI，the trinity of X，Y and Z，and is the first class of information conversion.

Fig.4 shows that the ontological information（denoted S）is applied to the input of the perception model and mapped to the corresponding syntactic information（denoted X）.Next，the pragmatic information（denoted Z）can be retrieved from the knowledge base，in which many X?Z pairs，｛X（i），Z（i）｝，are stored.When X is matched with X（i0），then Z（i0）is regarded as the pragmatic information corresponding to X.In case no math can be found，the equation can be used to find Z；

Z=Cor（X，G）（1）

where X and G are expressed as vectors；and Cor is the correla?tion operation.Because X and Z are now available，the seman?tic information Y can be inferred from：

Y=λ（X，Z）εS（2）

where S is the space of semantic information，andλis the logic operation mapping the pair of（X，Z）to Y in S.This means that Y is a subset of S when both X and Z are simultaneously valid. In other words，Y is determined by the joint conditions of X and Z（Fig.5）.

As a result，OI is converted into EI，which is the trinity of X，Y and Z，via the model in Fig.4.This technology is com?pletely feasible in practice.

Cognition technology can be implemented using the model in Fig.6，with which EI is converted to knowledge.This is the second class of information conversion.

According to the definitions of information and knowledge，information is the phenomenon in nature and knowledge is the essence in nature.Thus，knowledge can be established using inductive?type algorithms（Fig.6）.In Fig.6，comprehensive in?formation is another name for EI.

The technology for decision?making can be implemented us?ing the model in Fig.7，which converts EI to IS based on the support of the related knowledge and guided by the goal of problem solving given beforehand.

The model in Fig.7 is quite similar to that in Fig.6；howev?er，the big difference between them is the principal algorithms. In Fig.6，the algorithms are inductive type whereas in Fig.7 they are deductive type.

In AI theory and technology，the three classes of information conversion（IC）are the nucleus and novelty.Nevertheless，the sub?process of strategy execution is a kind of normal technolo?gy，converting the intelligent strategy into intelligent action，and therefore is not necessary to explain here anymore.

In sum，the key issue in AI theory and technology is to suc?cessively derive the EI（first class of IC）and knowledge（second?class IC）in order to deeply understand the object concerned and then produce an intelligent strategy（third?class IC）based on OI，EI，knowledge，and the goal of the system in order to regulate the object.

4 A Conceptualized Model of I2oT

The functioning ability of IoT is mainly limited by the perfor?mance of the embedded processor，which often fails to produce a strategy that is intelligent enough to deal with complex prob?lems.Therefore，strengthening the embedded processor is the key to improving IoT.

Because the intelligent strategy can be derived from the three classes of IC in AI，it is feasible to use these three class?es of IC to replace the embedded processor in IoT and trans?form IoT into I2oT，which is a much more intelligent version of IoT（Fig.8）.

Comparing the conceptual model in Fig.8 with that in Fig. 1，there is almost no difference between these two models ex?cept the replacement of embedded processor by the nucleus of AI technology，i.e.，first，second and third class of IC.This is the most effective way of transforming IoT into I2oT.

Because of the strong learning abilities offered by the three classes of IC in AI，the I2oT will be much more powerful than the conventional IoT and will be able to handle the complex problems mentioned in section 2.The intelligent strategy need?ed to tackle complex problems can，in principle，be derived from the three classes of IC based on OI，EI，relevant knowl?edge，and the designated goal for solving problems in a manner similar to humans.

5 Conclusion

In conclusion，the following two points are emphasized：

1）IoT will have to be capable of intelligently handling com?plex problems.This will be an increasingly serious chal? lenge.It is strongly recommended that the embedded proces?sor be replaced by the three classes of IC technology in AI. In this way，IoT will become I2OT and meet the demands of applications for economic and social development.

2）Broadly speaking，the real significance of AI is that it imple?ments the great law of information conversion and intelli?gence creation，according to which information is the means and intelligence creation is the purpose.This is the radical law that governs all information activities in the information era.Arguably，this law will be more significant than the law of energy conversion and conservation in physical science in industrial era.

［1］D.Kyriazis and T.Varvarigou，“Smart，autonomous and reliable internet of things，”in International Conference on Emerging Ubiquitous Systems and Perva?sive Networks，Niagara Falls，Canada，2013.doi：10.1016/j.procs.2013.09.059.

［2］W.C.McCulloch and W.Pitts，“A logic calculus of the ideas immanent in ner?vous activity，”Bulletin of Mathematical Biophysics，no.5，pp.115-133，1943.

［3］D.E.Rumelhart，Parallell Distributed Processing.MA，USA：MIT Press，1986.

［4］J.Hopfield，“Neural networks and physical systems with emergent collective computational abilities，”Proceedings of the National Academy of Sciences.，vol. 79，pp.2554-2558，1982.

［5］R.A.Brooks，“Intelligence without representation，”Artificial Intelligence，no. 47，pp.139-159，1991.

［6］R.A.Brooks，“Elephant cannot play chess，”Autonomous Robots，no.6，pp：3-15，1990.

［7］M.Wooldridge and N.Jennings，Intelligent Agents.Germany：springer-verlag，1995.

［8］N.Nilsson，Artificial Intelligence：A New Synthesis.San Mateo，USA：Morgan Kaufmann Publishers，1998.

［9］S.Russell and P.Norvig，Artificial Intelligence：A Modern Approach.Beijing，Chi?na：Pearson Education，Inc.，as Prentice Hall，Inc，and Tsinghua University Press，2006.

［10］Y.Zhong，Principles of Cognetics in Machine（in Chinese）.Beijing，China：Sci?ence Press，2007.

［11］Y.Zhong，Principles of Advanced Intelligence（in Chinese）.Beijing，China：Bei?jing Science Press，2014.

［12］Y.Zhong，“Principles of Information Conversion（in Chinese），”Science Bulle?tin，vol.58，no.14，pp.1300?13?6，2013.

Manuscript received：2015?04?11

Biographyraphy

Yixin Zhong（zyx@bupt.edu.cn）received the BS and MS degrees from Beijing Uni?versity of Posts and Telecommunications（BUPT）in 1962 and 1965.From 1979 to 1981，he was an academic visitor to the Department of Electrical Engineering，Impe?rial College of Science and Technology，London.He is now a professor in the De?partment of Intelligence Science，School of Computing，BUPT.From 1993 to 2005，he was an associated editor of IEEE Transactions on Neural Networks.From 2001 to 2002，he was chair of APNNA.From 2001 to 2010，he was president of the Chi?nese Association for Artificial Intelligence.From 2007 to 2009，he was vice?presi?dent of WFEO and the chair of WFEO?CIC.He has also been the general chair or program chair for a number of international conferences on communications，infor?mation science，and artificial intelligence.He is now the honorary president of Inter?national Society for Information Studies.

His research and teaching interests include information theory，neural networks，cognitive science，and artificial intelligence.

He received a number of national awards from the Chinese government and aca?demic organizations.He received the Outstanding Leadership Award from Interna?tional Neural Network Society（INNS）in 1994，President’s Award from Asia?Pacific Neural Network Assembly（APNNA）in 2002，Outstanding Contributor Award from World Federation of Engineering Organizations（WFEO）in 2010，National Out?standing Researcher and Professor in 2011，and Life Achievement Award from Chi?na Association for Artificial Intelligence in 2012.

He has published more than 450 papers and 18 books in related fields.