Neural computations underlying attention and expectation

Abstract

Our senses are inundated with an abundance of information at each instant. Attention and expectation are two cognitive processes that enable filtering this information to better navigate our environment. Importantly, attention filters information based on its relevance, whereas expectation filters information based on its likelihood. In this thesis, I study neural mechanisms and computations by which attention and expectation interact in space. For example, when crossing the road, the expectation of a potential collision guides our attention towards the side of the oncoming traffic. Given their close coupling, whether attention and expectation are mediated by the same, or distinct neural mechanisms remains actively researched. Moreover, common laboratory tasks – like the popular probabilistic cueing (or Posner cueing) task – employ expectation cueing to guide attention. As a result, whether such tasks measure attention, expectation or a mixture of both, remains an open question. Finally, how behavioral effects of attention and expectation emerge from their underlying neural signatures is also unknown. Here, I address these questions with a combination of psychophysics, human electrophysiology and Bayesian modeling. In the first Aim, I re-designed a psychophysical “dual-cueing” task (n=21 participants), inspired by earlier work, to separately quantify the effects of spatial attention and expectation cueing. With concurrent electroencephalography (EEG), I characterized neural signatures underlying each of these processes. Attention and expectation cueing modulated distinct psychophysical parameters, sensitivity and criterion, respectively. Moreover, neural metrics including steady state visually-evoked potential (SSVEP) power, alpha-band (8-12Hz) oscillation power, as well as those derived from representational similarity analysis (RSA), revealed markedly distinct effects of attention and expectation cueing. These results indicate that spatial attention and spatial expectation have distinct behavioral and neural underpinnings. In the second Aim, I investigated whether behavioral and neural effects of spatial probabilistic (Posner) cueing resemble those of attention, expectation or both. The same (n=21) participants who performed the “dual-cueing” task also performed a Posner cueing task. Posner cueing modulated both sensitivity and criterion. Yet, only sensitivity modulations by Posner cueing and those by attention cueing were correlated. Moreover, neural effects of Posner cueing, including SSVEP and alpha-band power modulation, resembled those of attention cueing. Neural networks trained to predict trial-level cue locations from EEG signals,showed that spectro-temporal feature representations of Posner cueing more closely resembled those of attention, rather than expectation, cueing. In other words, these results strongly suggest that Posner cueing recruits attention, and not expectation, mechanisms. In the final Aim, I explored how putative neural effects of attention and expectation could produce canonical patterns of psychophysical modulations. At the neural level, attention and expectation can either enhance target processing, suppress distractor processing, or both. With a novel Bayesian decision model, I showed that these two neural effects map distinctly onto distinct types of gain modulation on psychophysical functions. While target enhancement produces “contrast gain” – a threshold shift – distractor suppression yields “response gain” – a multiplicative scaling. I evaluated this model by designing two tasks with distinct types of stimuli (n=20 participants). In both tasks, manipulating stimulus statistics to mimic target enhancement or distractor suppression consistently yielded dissociable contrast and response gain effects, respectively, thereby validating the model’s predictions empirically. Overall, these experiments tease apart behavioral and neural effects of attention and expectation cueing. Additionally, they also show that neural computations associated with these processes – target enhancement and distractor suppression – produce distinct, canonical effects on psychophysical functions. More broadly, these results highlight the importance of distinguishing attention from expectation impairments when designing diagnoses and treatments for cognitive disorders, such as mild cognitive impairment, schizophrenia, autism or dementia.