Academia Sinica Logo

Abstract

Many living systems continuously sense and act to navigate their environments. Despite advances in behavioral tracking, quantitatively defining navigation strategies and linking them to internal states and computational functions remains challenging. In this talk, I will describe computational frameworks that unify the analysis of sensory navigation across scales, from animals to single cells. By viewing navigation as a dynamical system, we develop unsupervised methods to identify stereotyped behavioral strategies and to characterize internal states in an unbiased manner. We apply this framework to measurements of sensory navigation from bacteria, worms, flies, to machine-learning agents. In addition, we quantify the information flow between sensory signals and behavioral actions, proposing an information-theoretic metric for navigation performance derived from first principle. Finally, I will discuss progress towards experiments that probe the neural and biophysical mechanisms underlying experience-dependent navigation.