UCLA

GOAL: The goal of this dissertation is to investigate several factors influencing multisensory processing using a well-established Bayesian computational framework. Specifically, this project explores how systematic biases in human spatial perception, recalibration of the tendency to integrate sensory signals, and selective attention may (or may not) influence both likelihoods and priors in Bayesian models that can account for human performance quite well in a number of sensory tasks. These explorations are motivated by the following aims:

AIM 1: Exploring computational mechanisms underlying spatial biases in human visual, auditory, and audiovisual perception of space. Many investigations over the past twenty years have shown that human localizations along azimuth in both the visual and auditory modalities are often biased away from the true locations of the sensory stimulus. However, these investigations have often used small sample sizes and suboptimal paradigms to explore the nature of these spatial biases. In the first study of this dissertation, a dataset of unprecedented size (384 subjects) is used to explore the presence, magnitude, and direction of biases in humans’ visual, auditory, and audiovisual perception. Following this behavioral exploration, several versions of a Bayesian causal inference model are tested to see whether these spatial biases are best accounted for by biases in sensory representations of space (i.e. likelihoods), pre-existing biases for localization (i.e. priors), or a combination of these mechanisms. The best-fitting model from this investigation is then used as a computational tool in subsequent investigations.

AIM 2: Investigating how to recalibrate the tendency to integrate multisensory signals. It has previously been demonstrated in several multisensory investigations that human perception of auditory space can shift based on previous multisensory experiences, and that this effect can be computationally accounted for in a shift in the auditory likelihood distributions. However, little (if any) evidence has shown whether or not the a priori tendency to integrate multisensory signals can be recalibrated by previous sensory experiences. In regards to this second aim, six experiments are conducted to determine (1) whether the tendency to integrate can change to due simple, brief (10-minute) exposure tasks, and if so, (2) what the spatial and temporal rules are that the brain uses to update its tendency to bind.

AIM 3: Examining the effects of selective and divided attention on sensory integration. While the phenomenon of selective attention has been explored extensively by previous research, little consensus exists regarding the mechanism that enables enhanced perception for an attended stimulus. In this investigation, several experiments are conducted requiring subjects to allocate attention either selectively to one modality (vision or audition), or to divide attention across modalities, while performing either (1) a spatial task, or (2) a numerosity task. Data from both the spatial and temporal tasks are modeled to determine whether selective attention impacts the tendency to integrate multisensory stimuli, the sensory representations, or spatial priors. Leaving behind the antiquated idea of attention simply serving as a Bayesian prior, this series of studies demonstrates that attention primarily impacts sensory representations, and does not significantly impact the tendency to integrate multisensory stimuli.