Predictive Mobility Models based on Kth Markov Models

1
Predictive Mobility Models based on Kth Markov
Models

• Frédéric Lassabe, Damien Charlet, Philippe
  Canalda, Pascal Chatonnay and François Spies
• firstname.lastname_at_pu-pm.univ-fcomte.fr
• LIFC Montbéliard
• ICPS2006 - 06/28/2006

2
Introduction
Where am I ?
What can I do ?
Where do I go ?
What will I do ?
3
Related Work

• Sarukkai
• 1st Order Markov Model (MM)
• Pitkow and Pirolli
• Longest Repeating Sequences
• Kth MM (KMM) merged into All-KMM (AKMM)
• Deshpande and Karypis
• Light AKMM
• Selective MM

4
K-to-1 past

• AKMM shortcomings
• Error in iMM while jMM true (igtj)
• Apply to physical states
• Less connections ? Compact data
• Use every KMM for prediction
• Threshold
• Determines if a state is selected
• Can trigger complete or partial handoff (resp.
  prefetch)

5
Experiments
6
Conclusions Prospects

• Predictive power of Markov Models
• From web pages to mobility prediction
• Future works
• Analyse memory and computation costs
• Mobility Prediction and Multimedia Contents are
  real time, requires
• Time in the model
• Fast algorithm (distributed ?)
• Comparison with others models
• Bayesian Inference
• Neural Networks

7

• Thank you for your attention
• Any questions ?