Human behavior modeling allows many interesting applications. For example, in healthcare the information about a persons sleeping, eating and bathing behavior is very useful for monitoring the health status of elderly. Using a large number of small sensing nodes installed throughout the house, allows us to observe very simple actions. Pressure mats measure somebody lying in the bed, reed switches measure a door or cupboard opened, temperature sensors measure the use of the stove, etc. Recognizing activities from such sensor patterns is challenging because the interpretation of sensor readings is strongly context dependent. For example, opening a fridge can mean somebody is getting a drink or is getting some food to cook, it can even mean somebody is storing groceries they just bought. The context (i.e. the sensors that fire before and after the opening of the fridge) help in determining which activity is performed. However, it is difficult to model this context because the start and end point of an activity is not known. We use temporal probabilistic models to deal with these issues and have several ideas on how to extend these models to achieve better performance. We are looking for an enthusiastic AI master student with an interest in pattern recognition. We have real world datasets for evaluating the model and offer close supervision in solving a clearly defined problem.