Encoding of probability distributions of 3D estimates in mind and brain

项目摘要 最有影响力的生物视觉理论之一认为视觉感知是贝叶斯过程 推论。为了对外部世界做出推断，一个成功的视觉系统必须考虑到 考虑到神经计算的不确定性。在深度感知的特殊情况下，这一重点 项目中，贝叶斯模型假设不确定性明确地表示为定义的概率分布 可能对场景进行3D解释。只有这些知识才能集成多个3D源 实现贝叶斯最优的信息。与这一理论相容的大量数据来自研究 涉及深度辨别，其中确实发现知觉反应的变异性变得更小 随着更多的深度线索被添加到刺激中。在这里，人们质疑这些数据是否是证据表明 可变性源于神经噪声，代表3D估计的不确定性。相反，另一种理论是 提议的，这不需要这一代表。除了更加节俭，这一理论还可以 预测似乎证实了贝叶斯预测的相同发现。 这项探索性研究计划列出了这一新理论的两个可检验的预测，称为本征 约束(IC)，其中(1)大脑不表示3D属性上的概率分布，以及(2) 深度辨别任务的知觉变异性不能反映这些概率所编码的不确定性 分配。与贝叶斯理论不同，IC理论假设对不同3D刺激的反应 在大小上变化而不是感知的噪音。特别是，根据贝叶斯模型据称 具有不同的可靠性，因为IC模型会导致不同的感知收益。组合线索增加了 知觉增益和这个因素，而不是更高的精确度，提高了深度辨别任务的绩效。这 预测为IC模型提供了必要的解释力，以支持其作为3D理论的可行性 感知力。检验任何一种理论解释的有效性都将通过协同收集 行为和功能磁共振数据。首先，将确定两个-的仅可察觉的差异(JND)是否 间隔深度辨别任务测量的是与记忆保持相关的刺激可靠性或噪声。 根据第二种解释，是知觉收益决定了物理深度的变化 需要克服与此任务相关的噪音，与IC帐户一致。第二，一种功能磁共振成像技术 可以估计在神经种群活动中编码的概率分布的大小和噪声 提供关键的融合证据，证明存在(或不存在)3D不确定性的神经编码。在……里面 总而言之，这项研究项目将把视觉感知和视觉感知这两个独立的研究领域结合在一起 视觉短期记忆，研究利用行为和功能磁共振方法来解决 视觉科学中的基本问题。

Project Summary One of the most influential theories of biological vision considers visual perception as a process of Bayesian inference. In order to make inferences about the external world a successful visual system must take into account the uncertainty of neural computations. In the particular case of depth perception, the focus of this project, Bayesian models postulate that uncertainty is explicitly represented as probability distributions defined over possible 3D interpretations of a scene. Only this knowledge allows the integration of multiple sources of 3D information to achieve Bayesian optimality. A large body of data compatible with this theory comes from studies involving depth discrimination, where it is indeed found that variability in perceptual responses becomes smaller as more depth cues are added to a stimulus. Here it is questioned whether this data is evidence that behavioral variability stems from neural noise representing uncertainty of 3D estimates. Instead, an alternate theory is proposed, which does not require this representation. Beyond being more parsimonious, this theory can also predict the same findings that seem to confirm the Bayesian predictions. This exploratory research proposal lays out two testable predictions of this new theory, termed Intrinsic Constraint (IC), for which (1) the brain does not represent probability distributions over 3D properties and (2) perceptual variability in depth discrimination tasks does not reflect uncertainty encoded in these probability distributions. In contrast to the Bayesian account, the IC theory postulates that responses to different 3D stimuli vary in magnitude instead of perceptual noise. In particular, stimuli that according to Bayesian models allegedly have different reliabilities for the IC model elicit different perceptual gains. Combining cues increases the perceptual gain and this factor, not higher precision, enhances performance in depth discriminations tasks. This prediction gives the IC model the explanatory power necessary to support its viability as a theory of 3D perception. Testing the validity of either theoretical account will be achieved through the synergetic collection of behavioral and fMRI data. First, it will be determined whether the Just Noticeable Difference (JND) of a two- interval depth discrimination task measures stimulus reliability or noise associated with memory retention. According to this second interpretation, it is the perceptual gain that determines the changes in physical depth required to overcome this task related noise, in agreement with the IC account. Second, an fMRI technique that can estimate both the magnitude and noise of probability distributions encoded in neural population activity will provide critical converging evidence of the existence (or absence) of neural encoding of 3D uncertainty. In summary, this research project will bring together the two separate fields of research of visual perception and visual short-term memory, with investigations leveraging behavioral and fMRI methods for addressing a fundamental problem in vision science.