Summary, Project 5 (Models of computation in multi-regional circuits with thalamus in the middle) In close interplay with experiments from Projects 1-4, this computational project aims to understand how the dynamic interplay between cortex, thalamus and other subcortical structures underlies decision-making behaviors that depend on short-term memory and reward-dependent learning. We develop a theoretical framework for the multi-regional circuit mechanism underlying the generation and control of behavior-related activity modes in frontal cortical areas. In our framework cortical activity modes, corresponding to subspaces of neural population activity, are subject to control from thalamic nuclei (non-sensory, 'higher-order'), which in turn receive driving input from subcortical inputs. Moreover, we will study how subcortical inputs to the thalamus, such as from the midbrain and basal ganglia, interact in the thalamus to generate control strategies that organize sequences of cortical activity modes driving the behaviors required to solve a task. We explore the roles of cortical cell types that are activated by thalamocortical neurons in the generation and control of activity modes that are relevant for short-term memory maintenance and motor execution. We further study value-based decision making in a dynamic foraging task that leads to Herrnstein’s matching law of choice behavior. Our circuit model will implement two candidate mechanisms for updating and maintenance of action values, decision-making based on action values, as well as its transformation to behavioral choice and action. The model implements multiple loops involving cortex, subcortex and thalamus in the middle, all based on connectomic data (from Projects 1-3) and will guide multi-regional recordings (Projects 4). To conclude, computational research will be carried out in close coordination with anatomical, behavioral and neurophysiological experiments. Model predictions will be tested, and experimental results will shape the models. This collaboration will lead to foundational advances in our understanding of computations in multi- regional neural circuits.