As we interact with the world, we must decide what to do next based on previously acquired and incoming information. The study of perceptual decision-making uses highly controlled sensory stimuli and exploits known properties of sensory and motor systems to understand the processes that occur between sensation and action. Even these relatively simple decisions invoke operations like inference, integration of evidence, attention, appropriate action selection, and the assignment of levels of belief or confidence. Thus, the neurobiology of perceptual decision-making offers a tractable way of studying mechanisms that play a role in higher cognitive function. The controlled nature of perceptual decision-making tasks allows an experimenter to infer the latent processes that give rise to a decision. For example, many decisions are well-described by a process of bounded evidence accumulation, in which sensory evidence is temporally integrated until a terminating threshold is exceeded. This thesis improves our understanding of how these latent processes are implemented at the level of neurobiology.
After an introduction to perceptual decision-making in Chapter 1, Chapter 2 focuses on the behavioral observations that corroborate whether a subject’s decisions are governed by bounded evidence accumulation. Through simulations of multiple decision-making models, I show that several commonly accepted signatures of evidence accumulation are also predicted by models that do not posit evidence accumulation. I then dissect these models to uncover the features that underlie their mimicry of evidence accumulation. Using these insights, I designed a novel motion discrimination task that was able to better identify the decision strategies of human subjects.
In Chapter 3, I explore how the accumulation of evidence is instantiated by populations of neurons in the lateral intraparietal area (LIP) of the macaque monkey. Recordings from single LIP neurons averaged over many decisions have provided support that LIP represents the accumulation of noisy evidence over time, giving rise to diffusion dynamics. However, this diffusion-like signal has yet to be observed directly because of the inability to record from many neurons simultaneously. I used a new generation of recording technology—neuropixels probes optimized for use in primates—to record simultaneously from hundreds of LIP neurons, elucidating this signal for the first time. Through a variety of analyses, I show that the population’s representation of this signal depends on a small subset of neurons that have response fields that overlap the choice targets.
Finally, in Chapter 4, I discover a neural mechanism in the midbrain superior colliculus (SC) involved in terminating perceptual decisions. I show that trial-averaged activity in LIP and SC is qualitatively similar, but that single-trial dynamics in each area are distinct. Unlike LIP, SC fired large bursts of activity at the end of the decision, which were sometimes preceded by smaller bursts. Through simultaneous recordings, I uncover the aspects of the diffusion signal in LIP that are predictive of bursting in SC. These observations led me to hypothesize that bursts in SC are the product of a threshold computation involved in terminating the decision and generating the relevant motor response. I confirmed this hypothesis through focal inactivation of SC, which affected behavior and LIP activity in a way that is diagnostic of an impaired threshold mechanism. In total, this work improves our ability to identify the hidden, intermediate steps that underlie decisions and sheds light on their neural basis. All four chapters have been published or posted as separate manuscripts (Steinemann et al., 2022; Stine et al., 2020; Stine et al., 2022; Stine et al., 2019).
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/h0yy-zq26 |
| Date | January 2022 |
| Creators | Stine, Gabriel |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0026 seconds