*Pretalk by Daniel Durstewitz starts at 1.00 PM* In psychiatric research, we often encounter time series of different nature. These could be neural recordings obtained by functional magnetic resonance imaging of magnetoencephalography, physiological signals obtained, for instance, from the electrocardiogram, or behavioral choices collected during a cognitive experiment or via mobile health applications. The research group Computational Psychiatry at the Central Institute of Mental Health in Mannheim focuses on developing and applying both data driven and theory driven statistical models to explain the emergence of such time series. On the data driven side, our core approach consists of advancing models which aim at reconstructing the dynamical system (DS) hypothesized to generate the observed data. These models approximate the data generating DS by recurrent neural networks, and, in doing so, allow us to gain insights into attraction and repulsion dynamics of a system, as well as into disease informative phase transitions. They also allow to forecast future system states. On the theory driven side, we make use of behavioral (mostly) reinforcement learning models. These models are informed by theories on the types of algorithms humans solve to obtain a goal and can be modified according to the hypotheses an experimenter may have. Theory-driven models may be used to provide mechanistic insight into altered or maladaptive behavioral strategies, to identify co-varying neural substrates, or to define the experimental boundaries at which individuals will resort to healthier behavior. In this talk, I will show multiple examples of how we employ both approaches to gain an understanding on the mechanisms underlying choice, neural, and physiological responses, as well as to predict future responses.