Uncertainty in Applications

• System goal
  • Correctness
  • Efficiency
  • Scalability
• Difficulties
  • Uncertainty
  • Imprecision
  • Noise
  • Missing values
• Examples
  • Sampling accuracy
    • Values monitored is changing, and those stored in database may be obsolete
  • Uncertainty in satellite images
  • Measurement errors
    • GPS, indoor positioning, etc.
  • Uncertainty in repeated measurements
  • Text extraction
  • Data integration
  • Uncertain graphs
  • Criminal databases

Problems in Managing Uncertain Data

• Modeling data uncertainty
  • Capturing & representation
  • Simple v.s. complicated models
• Probabilistic queries
  • More difficult & expensive
• Data quality & cleaning
  • Measurement of the quality of data & quality of queries
  • Which & how to clean

Uncertainty Models

• Types
  • Attribute uncertainty (value uncertainty): I know it exists, but not sure about the value
    • Discrete attribute
      • Binary attribute
    • Continuous attribute
  • Tuple uncertainty (existential uncertainty): I know its value, but not sure if it exists at all
• Sources of uncertainty
  • Data staleness
    • Location-based tracking systems
      • Location estimation: last reported location + uncertainty model
  • Measurement errors
    • Value estimation: measured value + error estimation
      • Error modeled as e.g. Gaussian distribution
    • Dead reckoning: value bounded within [v-d, v+d]
      • Trade-off between data uncertainty & update frequency
  • Repeated measurements
    • Value estimation: distribution of a large number of of measurements
  • Data integration

Attribute Uncertainty

• $$f_i(x)$$: continuous, uniform, Gaussian, histogram, discrete, etc.

Uncertainty pdf/pmf

Depends on the application. If no information known, assume e.g. uniform pdf, or derive via time-series analysis from past data.

Tuple Uncertainty

X-Tuple Model

Integrating data from different sources, the values may vary for the same target.