explores Personal Ontology Learning, which is a new task to assist an individual to organize information items, such as documents, web pages, and email messages; to accomplish tasks; and to quickly access information. This thesis proposal presents two ontology learning frameworks, namely human-guided ontology learning and metric-based ontology learning. Both frameworks combine the strengths of existing pattern-based and clusteringbased ontology learning approaches, move beyond the limitation of traditional use of features, and incorporate a wide range of semantic evidence in ontology learning. Each framework provides a unique solution to achieve effective and flexible construction of ontologies. The two frameworks also differ in various ways with different emphases. The human-guided ontology learning framework addresses personalization and human computer interaction in the process of personal ontology learning. Periodic manual guidance provides training data for learning a distance metric, which is then used during automatic activities to further construct the ontology. A user study demonstrates that human-guided machine learning is able to generate ontologies with manually-built quality and with lower cost. It also shows that periodic manual guidance successfully directs machine learning towards personal preferences. The human-guided ontology learning framework is the first ontology learning framework to construct personalized and task-specific ontologies.

Surveillance video recording is becoming ubiquitous in daily life for public areas such as supermarkets, banks, and airports. The rate at which surveillance video is being generated has accelerated demand for machine understanding to enable better content-based search capabilities. Analyzing human activity is one of the key tasks to understand and search surveillance videos. In this thesis, we perform a comprehensive study on analyzing human activities from short term to long term and from simple to complicated activities in surveillance video achieves. A general, efficient and robust human activity recognition framework is proposed. We extract local descriptors at salient points from videos to represent human activities. The local descriptor is called Motion SIFT (MoSIFT) which explicitly augments appearance features with motion information. A quantization and classification framework then applies the descriptors to recognize activities of interest in surveillance

www.lti.cs.cmu.edu