Meistaranemi í Tölvunarfræði, Eva Ósk Gunnarsdóttir, mun verja MSc ritgerð sína (60 ECTS) þann 18.janúar næstkomandi kl. 12:00-13:00 í stofu V 103.

Title: Applying Binary Decision Diagrams to Learn Hidden Markov Models

Abstract:

In this thesis, we present EM-BDD, an algorithm for learning parameters of Hidden Markov Models, by building upon the Baum-Welch (BW) algorithm’s methodology. EM-BDD utilises the Forward-Backward procedure, a cornerstone of BW, adapting it to operate on Binary Decision Diagrams (BDDs). The time and memory complexity of the algorithm is contingent on the size of the BDD, highlighting that the BDD’s size significantly depends on the variable ordering (a problem known to be NP-complete). Preliminary experiments showed that the BDD size grows exponentially with the size of the input, giving the EM-BDD algorithm the same time complexity as the BW algorithm on average. However, this does not encompass the best-case scenario, which could potentially exhibit a more favourable time complexity

Keywords: Baum-Welch algorithm, Forward-Backward procedure, Binary Decision Diagrams, Hidden Markov Models & Expectation-Maximization algorithm.

Committee:

• Anna Ingólfsdóttir, Supervisor, Professor, Reykjavík University, Iceland
• Giovanni Bacci, Examiner, Associate Professor, Aalborg University, Denmark
• Luca Aceto, Examiner, Professor, Reykjavík University, Iceland