1
Representing Temporal Knowledge for Case-Based
Prediction

• Martha Dørum Jære, Agnar Aamodt, Pål Skalle

2
Introduction

• Current CBR snap-shots in time, temporal
  relations ignored or handeled explisit within
  reasoning algorithms
• Real world context (more interactive and
  user-transparent)

3
Creek

• integrates cases with general domain konwledge
  within a single semantic network
• feature and feature value -gt concept in semantic
  network
• Interliked with other consept, semantic relations
  specified in general domain model
• General domain knowledge model based reasoning
  support to the CBR processes Retrieve, Reuse and
  Retain

4
Overview

• Related research
• Summary of James Allens temporal intervals
• Introduces problem of predicting unwanted events
  in an industiral process
• Temporal representation in system
• How representation is utilized for matching of
  temporal intervals

5
Overview

• Related research
• Summary of James Allens temporal intervals
• Introduces problem of predicting unwanted events
  in an industiral process
• Temporal representation in system
• How representation is utilized for matching of
  temporal intervals

6
Related research

• Early AI research on temporal reasoning make
  distinction between point-based (instans-based)
  and interval-based (periode-based)(Allen)
• Jaczynski and Trousse Time-extended situations
• Mendelez supervicing and controlling sequencing
  of process steps that have to fulfill certain
  conditions

7
Related research (2)

• Hansen weather prediction
• Branting and Hastings pest management, temporal
  projection
• McLaren Ashley temporal intervals, engineering
  ethics system

8
Hypothesis

• Large and complex data
• Explanatory reasoning methodes underlying the CBR
  apporach
• Strongly indicate that a qualitative,
  interval-based framework for temporal reasoning
  is preferrable

?
9
Overview

• Related research
• Summary of James Allens temporal intervals
• Introduces problem of predicting unwanted events
  in an industiral process
• Temporal representation in system
• How representation is utilized for matching of
  temporal intervals

10
Allens temporal intervals

• Interval-based temporal logic
• Intervals decomposable
• Intervals may be open or closed
• Intervals hierarchy connected by temporal
  relations
• During hierachy propostions inhereted
• 13 ways ordered pair of intervals can be related
  (mutually exclusive temporal rel.)

11
Allens 13 ways
12
Allens temporal intervals(2)

• Temporal network, transitivity rule
• Generalization method using reference intervals

13
Overview

• Related research
• Summary of James Allens temporal intervals
• Introduces problem of predicting unwanted events
  in an industiral process
• Temporal representation in system
• How representation is utilized for matching of
  temporal intervals

14
Prediction of unwanted events

• Oil drilling domain
• Stuck pipe situation
• Alert state
• Alarm state

15
Overview

• Related research
• Summary of James Allens temporal intervals
• Introduces problem of predicting unwanted events
  in an industiral process
• Temporal representation in system
• How representation is utilized for matching of
  temporal intervals

16
Temporal representation in Creek

• Allens approach
• Intervals stored as temporal relationships inside
  cases
• Cases restrict computational complexity
• Transitivity
• Case explanations

17
Temporal representation in Creek(2)

• Two intervals added

• For every new interval that is added to the
  network
• Create a lthas intervalgt relationship
• Create lthas findinggt relationships
• Create ltTemporal Relationgt relationships
• Infer new ltTemporal Relationgt relationships

18
Temporal representation in Creek(3)
19
Overview

• Related research
• Summary of James Allens temporal intervals
• Introduces problem of predicting unwanted events
  in an industiral process
• Temporal representation in system
• How representation is utilized for matching of
  temporal intervals

20
Temporal Paths Dynamic Ordering

• Original
• Activation strength
• Explanation strength
• Matching strength

• Temporal similarity matching
• Temporal path strength

21
Temporal Paths Dynamic Ordering (2)

• Dynamic ordering algorithm
• Find first interval in IC and CC
• Check intervalIC and intervalCC for matching or
  explainable findings
• If match - Update temporal path strength
• Check getSameTimeIntervals for new information
  and special situations
• If special situations - Perform action
• getNextInterval from CC and IC
• Unless getNextInterval is empty - Go to (2)
• Return temporal path strength

22
Example Prediction

• Oil-well drilling
• Highlights
• Retrieving similar cases (matching strength above
  treshold)
• Compare -gt temporal path stregth
• i.e. alerts

23
Conclusion

• Support prediction of events for ind. processes
• Allens temporal intervals incorporated into
  Creek

I
24
Conclusion (2)

• Intervals-gtcloser to human expert think
• Integration into model based reasoning system
  component

25
Conclusion (3)

• -
• One fixed layer of intervals
• System Raw data -gt qualitative changes
• Many processes too complex

26
Discussion

• Hypotheses ?
• How represent time intervalls in cases? (When
  having to monitore over time?)
• Continous matching? Or treshold/event driven?