Detecting changes in streaming data using adaptive estimation

Dean Bodenham, ETH Zürich

( joint work with Niall Adams)

Outline

  • Change detection in streaming data
  • Applications
  • Adaptive estimation using forgetting factors
  • Extensions

Streaming data

Data streams

A data stream is a sequence of random observations \[x_1, x_2, \dots, x_N, \dots \]

  • observed sequentially
  • potentially unending
    • too many observations to store and analyse later
  • observations arrive at a high frequency
  • underlying distributions are often unknown
  • unpredictable when and how changes occur

Change detection

Change detection in streaming data