Context Prediction Based on Branch Prediction Methods

Abstract:

Ubiquitous systems use context information to adapt appliance behavior to human needs. Even more convenience is reached if the appliance foresees the user s desires and acts proactively. This paper focuses on context prediction based on previous behavior patterns. The proposed prediction algorithms originate in branch prediction techniques of current high-performance microprocessors which are transformed to handle context prediction. We propose and evaluate the onelevel one-state, two-state, and multiple-state predictors, and the two-level two-state predictors with local and global rst-level histories. Evaluation is performed by simulating the predictors with behavior patterns of people walking through a building as workload. The evaluations show that the proposed context predictors perform well but exhibit differences in training and retraining speed and in their ability to learn complex patterns.

Downloads:

• Technical Report as pdf - (2003_14_pet_pdf.pdf, 768 KB)

Links:

• Technical Report in OPUS