Reasoning about Uncertain Contexts in Pervasive Computing Environments

1
Reasoning about Uncertain Contextsin Pervasive
Computing Environments

• 2004.10.27
• ? ??

2
Overview

• Introduction
• Context Infrastructure - Gaia
• Uncertainty Model
• ConChat
• Reasoning Mechanisms
• Examples
• Conclusion

3
Introduction

• Context-Aware systems can't always identifythe
  context precisely
• Need support for handling uncertainty
• Uncertainty Model
• Context Predicate
• Confidence Value
• Reasoning Mechanisms
• Probabilistic Logic
• Fuzzy Logic
• Bayesian Networks
• Context Infrastructure - Gaia

4
Context Infrastructure - Gaia

• Context Providers
• Sensors or other data sourcesof context
  information
• Context Synthesizers
• Get sensed contextsfrom various Context
  Providers
• Deduce higher-level or abstract contexts
• Provide deduced contexts to other agents
• Context Consumers
• Entities (Context-Aware applications)
• Get different types of contexts from Context
  Providers or Context Synthesizers

5
Context Infrastructure - Gaia

• Context Provider Lookup Service
• Enables Context Providers to advertise what they
  offer
• Enables agents to find appropriate Context
  Providers
• Context History Service
• Lets agents query for past contexts
• Ontology Server
• Maintains the ontologies

6
Uncertainty Model

• Context Predicate
• Simple, uniform representation for different
  kinds of contexts
• Subject-Object(ContextType (ltSubjectgt,
  ltObjectgt))or Subject-Verb-Object(ContextType(ltSu
  bjectgt, ltVerbgt, ltObject gt))
• ex. location(jeff, in, room 3105)
  activity(room 3102, meeting)
• Comparison operator(, gt, lt), Boolean
  operations(And(?), Or(?), Not()) ,
  Quantification(?, ?) is possible
• ex. Sound Level(dB, gt, 60) Environment
  Lighting(Room 3234, Is, Off) ? Environment
  Lighting(Room 3234, Is, Dim) Environment
  Lighting(Room 3234, Is, Off ? Dim) ?Location
  Y Context(Location, Chris, In, Y)

7
Uncertainty Model

• Confidence Value
• Measures the probability or the membership value
  of the event
• Attaching between 0 and 1 to predicates
• ex. prob(location(carol, in, room 3233)) 0.5

8
ConChat

• A Context-Aware chat program
• Let users query their chat partner's context
  through a side channel
• Context Model
• Context Predicate
• Architecture for
  Context-Aware Chat

9
ConChat

• Sharing contextual information
• Party location
• Number of other people in the room
• The identities of the other people in the room
• Room temperature, light, and sound
• Other applications and devices running in the
  room
• User's mood (such as happy, sad, or excited)
• User's status (such as "on the phone" or "out to
  lunch")
• Activity in the room (such as a meeting, lecture,
  or presentation)
• Activity
• first-order predicate calculus that it evaluates
  to determine the room's activity.
• ex. people(Room 2401, gt, 3) ?
  Application(PowerPoint, Is, Running) gt Room
  Activity(2401, Is, Presentation) people(Room
  2401, gt, 3) ? NOT Application(PowerPoint, Is,
  Running) gt RoomActivity(2401, Is, Meeting)

10
ConChat

• Semantic conflicts
• Naming conflicts
• Confounding conflicts
• Scaling conflicts
• ex. AppearsText("Football") ?
  Location-Country(Sender, In, Canada ?USA) ?
  Location-Country(Receiver, In, France ? Germany ?
  UK) ? Tag-With(American Football)
  Currency(Sender, Is, X) ? AppearsMoney(X, N) ?
  Currency(Receiver, Is, Y) ? (X ? Y) ?
  TagWith(ConvertCurrency(N, X, Y) )

11
ConChat

• ConChat user
  interface

12
Reasoning Mechanisms

• Probabilistic Logic
• Lets us write rules that reason about events'
  probabilities in terms of the probabilities of
  other related events
• ex. prob(X, Y, union, P) - prob(X, Q), prob(Y,
  R), disjoint(X, Y), (P is Q R)
• ? Pr(X ?Y) Pr(X) Pr(Y) if X and Y are
  disjoint events
• Fuzzy Logic
• Capturing and representing imprecise notions
  (such as "tall", "trustworthy" and "confidence")
  and reasoning about them
• Confidence values represent degrees of membership

13
Reasoning Mechanisms

• Bayesian Networks
• Directed Acyclic Graphs
• Each value corresponds to a certain context
  predicate
• Rootnodes representthe information to be deduced
• Leaves are sensed information
• Intermediate nodes are subgoals that are
  helpfulto the deduction process

14
Structure of an entity
15
Examples

• Uncertainty in sensing context
• Two factors contribute to uncertainty in a
  person's location
• The uncertainty of whether the badge is actually
  in the room
• Using a sensor fusion procedure
• The uncertainty of whether the people actually
  carry the badge
• Assign a probability value to the event of a
  person actually having the badge in hand
• ex. Prob (location(X, in, Y)) Prob (device
  associated with person is in Y) Prob (device
  is actually with person X)
• ? Use in Authentication

16
Examples

• Inferred context
• Authentication service
• Confidence Value V1, V2,..., V
• ? Vnet 1 - (1 - V1)(1 - V2) ...(1 - Vn)
• Room activity
• Bayesian inferencing algorithm
• Calculates the conditional probability
  distribution of the Activity node given the
  states of the leaf nodes
• ex. activity(room 2401, meeting) ? Confidence
  value 0.6 activity(room 2401, presentation) ?
  Confidence value 0.4 activity(room 2401,
  idle) ? Confidence value 0.9

17
Examples

• Applications' use of uncertain context
  information
• Access Control
• ex. CanAccess(P, display) -
  ConfidenceLevel(authenticated(P), C), C gt 0.7,  
  Prob(activity(2401,cs 101 presentation), Y), Y gt
  0.8, possessRole(P, presenter)

18
Conclusion

• Key Factor
• Uncertainty Model
• Context Predicate
• Confidence Value
• Reasoning Mechanisms
• Probabilistic Logic
• Fuzzy Logic
• Bayesian Networks
• Meaningful in understanding Context-Aware