Proceedings of the Annual Meeting of the Cognitive Science Society

We demonstrate that inference-based goal-directed behavior can be done by utilizing the temporal gradients in re-current neural network (RNN). The RNN learns a dynamic sensorimotor forward model. Once the RNN is trained, it can beused to execute active-inference-based, goal-directed policy optimization. The internal neural activities of the trained RNNessentially model the predictive state of the controlled entity. The implemented optimization process projects the neural activ-ities into the future via the RNN recurrences following a tentative sequence of motor commands (encoded in neurons akin torecurrent parametric biases). This sequence is adapted by back-projecting the error between the forward-projected hypotheticalstates and desired (goal-like) system states onto the motor commands. Few cycles of forward projection and goal-based errorbackpropagation yield the sequences of motor commands that control the dynamical systems. As an example, we show that atrained RNN model can be used to effectively control a quadrocopter-like system.