The information received through our senses is inherently probabilistic, and one of the main tasks faced by the brain is to construct an accurate representation of the world in spite of this uncertainty. This problem is particularly relevant when considering the integration of multiple sensory cues, since the uncertainty associated with each cue can vary rapidly and unpredictably. Recent psychophysical studies have shown that human observers combine cues by weighting them in proportion to their reliability, consistent with statistically optimal schemes derived from Bayesian probability theory. Remarkably, because cue reliability was varied randomly across trials, the setting of the weights must occur on the time scale of a single stimulus presentation.

The neural basis of cue re-weighting remains unknown, in part due to the lack of a suitable animal model system for simultaneous behavioral and neurophysiological measurements in the context of cue integration. We have developed a psychophysical paradigm in monkeys in which they are trained to report their self-motion direction (heading), using visual cues (optic flow), vestibular cues (inertial motion), or a combination of both. On a subset of trials, a small conflict angle was introduced between the visual and vestibular heading trajectories, and the relative reliability of the cues was varied by interleaving different levels of visual motion coherence.

We found that monkeys can dynamically re-weight cues according to their reliability, the first such demonstration in a species other than humans. During the task, we recorded single units from area MSTd, a region strongly implicated in the processing of visual and vestibular self-motion cues. We used ROC analysis to quantify the behavior of an ideal observer performing the same task as the animal but using only the firing rate of the neuron. Preliminary results suggest that MSTd neurons exhibit dynamic cue re-weighting with changes in reliability, analogous to the monkeys’ behavior. To our knowledge, this result provides the first direct evidence of a neural implementation of Bayesian inference in multisensory processing.