The SAIS board is happy to award Mattias Tiger the 2015 SAIS Master Thesis Award for the thesis ”Unsupervised Spatio-Temporal Activity Learning and Recognition in a Stream Processing Framework” with the following motivation:
“The award goes to Mattias Tiger for a thesis, which presents extensive work that required substantial theoretical understanding, and which has resulted in a method with high potential for becoming used. Tiger has developed a novel trajectory clustering approach for unsupervised on-line learning of spatio-temporal activities of individual objects that can be used for both classifying ongoing activities and predict future activities. The representation learned includes both a continuous part representing the continuous trajectory of an object and a discrete part representing the transitions between primitive activities. The foundation for the work is a new approach to sparse Gaussian Processes suitable for representing trajectories. The problem is highly relevant for the AI field, not only for activity classification, but also for more generic non-parametric modelling of time series for several different applications. The work has so far resulted in two publications and in receiving a PhD student grant from the national graduate school in computer science (CUGS). The solution has scientific height, is well motivated and clearly described. The report is clear and pedagogical written.”
The project was supervised by Daniel de Leng and Fredrik Heintz, both at Linköping University.