Continual learning of interval-based temporal patterns from event streams

This research focuses simultaneously on several challenges in machine learning; continual learning of time-interval patterns, in an unsupervised manner, from a streaming input of symbolic time sequences. The key contributions of this thesis are a model of Temporal Constraint Patterns (TCPs) and two learning algorithms. The two proposed algorithms perform unsupervised, continual learning from event streams, which is achieved by incrementally adding patterns to TCP model. Experiments were performed on several benchmark datasets and compared to existing approaches. The results show that the approach is capable of forming accurate model representations and demonstrate the key aspects of the system design.