Learning-automaton-based online discovery and tracking of spatiotemporal event patterns

Area of Science:

• Ubiquitous Computing
• Pattern Recognition
• Machine Learning

Background:

• Event notification systems in ubiquitous computing generate excessive, often redundant, data.
• This redundancy can make event sharing obtrusive rather than helpful.
• Discovering spatiotemporal patterns in noisy event sequences is a significant challenge.

Purpose of the Study:

• To develop a new scheme for discovering and tracking noisy spatiotemporal event patterns.
• To suppress recurring patterns and identify novel events in real-time.
• To improve the efficiency and reduce obtrusiveness of event notification systems.

Main Methods:

• A novel scheme based on maintaining hypotheses of spatiotemporal event patterns.
• Utilizing a dedicated Spatiotemporal Pattern Learning Automaton (STPLA) for each hypothesis.
• Employing a real-time guided random walk to infer hypothesis correctness and adaptively suppress redundant information.

Main Results:

• The STPLA scheme demonstrates superior convergence and adaptation speed in simulations.
• It effectively operates with noisy data, handling both event inclusion and omission errors.
• Empirical comparisons confirm its superiority over state-of-the-art approaches, particularly in robustness to noise.

Conclusions:

• The STPLA scheme is computationally efficient with a minimal memory footprint and is ergodic for adaptation.
• Its robustness to noise (inclusion and omission) is a unique and valuable property.
• The STPLA scheme is ideal for enhancing event notification and sharing systems by adaptively suppressing redundant information.