Berlin 2018 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

SOE: Fachverband Physik sozio-ökonomischer Systeme

SOE 22: Focus Session: Computational Social Science

SOE 22.3: Talk

Thursday, March 15, 2018, 16:45–17:00, MA 001

Comparison of Baum Welch Algorithm and Simulated Annealing as training algorithms for Hidden Markov Models — •Kim Schmidt and Karl Heinz Hoffmann — TU Chemnitz, Institut für Physik, 09107 Chemnitz, Germany

In some situations, such as an old lady at a new ticket machine or a driver in a highly automated vehicle, it is very important to identify helplessness to offer assistance and avoid or reduce frustration. We intend to use facial expression, gestures, and voice or lip movement to train a Hidden Markov Model (HMM) that should identify conditions as joy, frustration and helplessness. The Baum Welch Algorithm (BW) is the common training algorithm with the drawback of getting stuck in local minima. An alternative algorithm can be Simulated Annealing (SA) that can overcome local minima and thus it can end in a better solution. In particular we focus on comparing both algorithms for a varied complexity of exemplary HMMs.