Relationship Classification of Object to Object communications in - - PowerPoint PPT Presentation

D R P A T D O O D Y , D I R E C T O R O F T H E C E N T R E F O R I N N O V A T I O N I N D I S T R I B U T E D S Y S T E M S ( C I D S ) I N S T I T U T E O F T E C H N O L O G Y T R A L E E M R A N D R E W S H I E L D S I R C S E T F U N D E D R E S E A R C H E R C E N T R E F O R I N N O V A T I O N I N D I S T R I B U T E D S Y S T E M S I N S T I T U T E O F T E C H N O L O G Y T R A L E E

Relationship Classification of Object to Object communications in the Internet of Things using Reality Mining

1

Overview

 Reality Mining?  Applications  Our Projects  Reality mining Applied to IoT  Conclusions

2

Reality Mining

 “…The collection and analysis of machine-sensed

environmental data pertaining to human social behaviour …”

 .. Extracting information from real world sensor

data

 With the goal of identifying predictable patterns of behaviour.

 It was declared to be one of the "10 technologies

most likely to change the way we live" by Technology Review Magazine.

3

Reality Mining

4

Enabled Applications

Environmental Monitoring - Noisetube

5

Real-time Traffic Monitoring

6

Mobile Millennium, UC Berkeley

7

Citysense

 Shows the overall activity

level of the city,

 Highlights top activity

hotspots in real-time.

 Then it links to Yelp and

Google to show what venues are operating at those locations.

8

Market Players

9

Clustering

 A common property of human social networks are

cliques, circles of friends or acquaintances

 This inherent tendency to cluster is quantified by the

clustering coefficient [Watts and Strogatz (1998)].

 Nodes that are clustered together can easily

communicate with each other.

 Previous research in this area (Ghiasi, et al. 2002)

has studied the theoretical aspects of this problem

 Applications to energy optimisation.

10

Small-world networks

 Objects may only use knowledge of their

• wn acquaintances, to collectively construct

paths to the target.

 “six degrees of separation” found by the

social psychologist Stanley Milgram

 This raises a fundamental question

 Why should this type of decentralised routing so

effective?

11

Algorithm Considerations

 Algorithms must take into consideration the

characteristics of networks

Energy, Computation constraints, Network dynamics, and faults.  K-Nearest Neighbor Algorithm  ART1

 Weighted Regression  Case-based reasoning

12

Clustering - Voronoi Diagram

 Decision surface formed by the training examples

13

Regression - Bayesian MLP

 Several techniques can be

used in movement predictions

 Artificial Neural Networks,  Bayesian Belief Networks  Hidden Markov Chains  Dynamic Belief Networks  Each technique has its

advantages and disadvantages.

 Typically use a hybrid

model

14

Degree Distribution

 Nodes in a network typically do not all have the same

number of links, or degree.

 For a large number of networks

 The World Wide Web [Albert et al. (1999)],  The internet [Faloutsos et al. (1999)]  metabolic networks [Jeong et al. (2000)],

 The work listed above assumes a static network

topology

 Complex IoT networks will continuously changing

• ver time.

15

CIDS Research

 Telecommunication Caching  Reality Mining applied to mobile networks  Classifying user groups  Predicting network usage patterns  Using Neural Network and other techniques

16

Reality Mining in mobile networks

17

Reality Mining in transport networks

 It is possible to infer an

individual’s

 Daily commute to work  Amount of time spent at work, at

home and traveling

 Allowing individuals to make

better traveling decisions.

 Provides information which will

be used to proactively manage the transportation network.

 Several clustering algorithms

base on artificial intelligence and statistical analysis will need to be considered and evaluated

 Adaptive Resonance Theory,  Eigenbehaviours

18

Enabling technologies

 The widespread adoption of the Internet of Things will take

time

 First: in order to connect everyday objects item identification

is crucial.

 Radio-frequency identification (RFID) offers this functionality.

 Second: the ability to detect changes in the physical status of

things, using sensor technologies.

 Embedded intelligence in the things themselves can further enhance the

power of the network

 Third: advances in miniaturisation and nanotechnology mean

that smaller and smaller things will have the ability to interact and connect.

 A combination of all of these developments will create an

Internet of Things that connects the world’s objects in both a sensory and an intelligent manner.

19

Reality mining Applied to IoT

 Data mining as applied to “business intelligence”

applications may play a role

 Techniques currently applied to understanding

human behaviour and interactions may be applicable to IoT systems.

 Reality Mining is one such technique.

20

Mining the IoT Social Network

 Relationships between smart object in an IoT

network

 May have similar properties to humans interacting in a social

environment.

 When smart objects participate in context-aware

applications

 Changes in their real-world environment impact on underlying

networking structures.

 Vast amounts of data being generated by smart

• bjects

 Modelled and applied to complex IoT networks

21

Mining the IoT Social Network

 Randomness (entropy)  Inherent in human social networks  Entropy of a smart object may be used as a metric

22

Mining the IoT Social Network

 Dyadic Inference.  Human social networks respond to surrounding

social environment

 Smart objects may exhibit similar dyadic properties.  From these properties it is possible to infer

 Relationships between multiple smart objects  based on patterns in proximity data.  Smart objects related in such a manner may responds to

environmental stimuli

23

Why do we care?

 Social Science

 Social Network Analysis  Behavioural Modelling  Human Mobility

 Systems Research

 Transportation  Environmental Modelling  Healthcare

24

Applications

 User-Generated Content is a core aspect of the Web

 online social networks  Blogs  wikis,  Forums

 One of the most successful services allowing this is

Twitter:

 Possibility is the development of Things-Generated

Content where Things (instead of human beings) are provided with "tweet-capabilities"

25

Challenges

 Large Datasets

 Wal-Mart: 100-400 GB/day of RFID data  CERN LHC: 40 TB/day

 Storage is cheap!  Stream data mining

26

Challenges

 Abstraction

 Low level details  Parallelism  Task distribution  Load balancing  Fault tolerance

 Google MapReduce

 Framework introduced by Google to support distributed

computing on large data sets on clusters of computers

27

Challenges

 Privacy

 Right to possess your data  Control the use of your data  Right to distribute or dispose of your data

28

Conclusions

 The Internet of Things has great promise  Business, policy, and technical challenges must be

tackled before these systems are widely embraced.

 Early adopters will need to prove that the new sensor

driven business models create superior value.

 Industry groups and government regulators should

study rules on data privacy and data security, particularly for uses that touch on sensitive consumer information.

 Software to aggregate and analyse data, must improve

to the point where huge volumes of data can be absorbed by human decision makers or synthesised to guide automated systems more appropriately.

29

Conclusions

 On the technology side, the cost of sensors and

actuators must fall to levels that will spark widespread use.

 Networking technologies and the standards that

support them must evolve to the point where data can flow freely among sensors, computers, and actuators.

 Within companies, big changes in information patterns

will have implications for organisational structures, as well as for the way decisions are made, operations are managed, and processes are conceived.

 Product development, for example, will need to reflect

far greater possibilities for capturing and analysing information.

30

Q&A

31