Vestibular Cognition: A Computational Approach to Sensory Inference and Cognition

This dissertation is a compilation of publications and manuscripts that seek to advance the burgeoning field of vestibular cognition from two perspectives: (i) by developing a computational framework through which we can connect high-level cognitive capacities to perception of self-motion, and (ii) by investigating whether the perception of self- motion is itself cognitively penetrable, i.e. whether it can be influenced by cognition. In the first manuscript, we provide a formal account of the computations that are similar between self-motion perception, and imagined self-motion, which is conceptualized as a simulation of a dynamical system. This approach aims to provide novel insights into mental imagery, and introduces our sense of self-motion as a suitable sensory modality in which to investigate the connections between cognition and perception. This is largely due to the wealth of computational models that describe the vestibular sensory system in terms of probabilistic inference; this, in turn is due to the comparative simplicity of the peripheral vestibular sensors. In the second manuscript, we discuss how cognitive training may be beneficial to patients suffering bilateral loss of their vestibular sensors, and we seek to understand this in terms of computations that may enable patients to compensate for their loss of sensory input. In the third manuscript, we examine the effect of prior knowledge about the direction of passive self-motion on self-motion decision making. In this empirical study, we sought to determine which cognitive processes are affected by prior knowledge, using a diffusion decision model to analyze subjects’ choices and response times. Although not conclusive, we provide evidence that subjects incorporate their prior knowledge by both biasing their decision-making process and by changing their rate of evidence accumulation. The fourth manuscript is a perspective paper, in which we claim that, in order to understand mental imagery, it is not sufficient to focus merely on neural resources that are shared with perception, but that it is necessary to focus on the computations that are common to both. As such, this paper is a precursor to the first manuscript, which constitutes this thesis’ main contribution