Tactile perception as probabilistic inference

人类认为感知外部世界的能力是理所当然的，但神经系统是如何完成这一非凡壮举的呢？我们的感觉器官将物理刺激转化为神经元反应的方法，以及大脑中感觉神经元的刺激诱发的放电模式，相对来说已经被很好地理解了。对于感知的奥秘，人们了解得很少：在感觉激活之后，大脑是如何重建物质世界的图像的？我们的研究将解决这个重要的神经科学问题，具体涉及触觉。******大脑中感觉神经元对重复的、相同的刺激的反应是高度可变的。因此，感知本质上是不确定的，并且涉及到从其可变效应（神经活动模式）推断原因（物理刺激）的具有挑战性的“逆问题”。这个具有挑战性的问题通过贝叶斯概率推理得到了最佳解决。******理想的贝叶斯大脑根据先前的经验来评估不确定的感觉数据。具体来说，它考虑了两个概率：1)观察到的神经活动的概率，给定每个可能引起它的刺激（“可能性”），2)每个刺激发生的概率，基于先前的经验（“先验”）。这两个元素结合起来计算每个刺激的最终概率（“后验”）。贝叶斯模型通常假设最终概率最高的刺激作为知觉呈现给意识。******贝叶斯模型在视觉感知的研究中应用较多，但在触觉感知的研究中应用较少。在一系列的研究中，我们将把人类的触觉感知与贝叶斯模型的预测进行比较。我们的研究将探讨期望（先前的经验）、适应（对持续刺激的神经反应的减少）、多指整合（用几个手指同时感觉物体的感知）和触觉探索（主动触摸物体识别）对人类通过触摸感知能力的影响。研究结果将表明触觉感知是否在贝叶斯意义上是最优的。******这项工作将具有重要的理论和实践意义。这项工作将加深对感知处理的基本科学理解，也可能为下一代触觉图形显示、触觉虚拟现实系统和机器人技术的构建提供信息。为了设计一个成功的虚拟现实系统，科学家和工程师必须首先了解最终用户如何处理感官输入以达到感知推断。同样，为了创造类人机器人系统，科学家和工程师必须首先了解人类的感知处理。我们的研究将使这种理解成为可能，从而使加拿大的科学和工程受益

Humans take for granted our ability to perceive the external world, but how does the nervous system accomplish this remarkable feat? The means by which our sensory organs convert physical stimuli into neuronal responses are relatively well understood, as are the stimulus-evoked firing patterns of sensory neurons in the brain. Much less understood is the mystery of perception: following sensory activation, how does the brain reconstruct an image of the physical world? Our research will address this important neuroscientific question with specific reference to the sense of touch.******The response of a sensory neuron in the brain to a repeated, identical stimulus is highly variable. Perception is thus inherently uncertain, and involves the challenging "inverse problem" of inferring a cause (the physical stimulus) from its variable effects (patterns of neural activity). This challenging problem is optimally approached via Bayesian probabilistic inference. ******The ideal Bayesian brain evaluates uncertain sensory data in light of prior experience. Specifically, it considers two probabilities: 1) the probability of the observed neural activity, given each stimulus that could have caused it (the "likelihood"), and 2) the probability of occurrence of each stimulus, based on previous experience (the "prior"). These two elements are combined to calculate the final probability of each stimulus (the "posterior"). Bayesian models typically assume that the stimulus with highest final probability is presented to consciousness as the percept.******Bayesian models have been applied frequently in studies of visual perception but only rarely in studies of tactile perception. In a series of studies, we will compare human tactile perception to the predictions from Bayesian models. Our studies will probe the effects of expectation (prior experience), adaptation (the reduction in neural response to a sustained stimulus), multi-finger integration (perception of objects felt simultaneously with several fingers), and haptic exploration (active touch for object recognition) on the ability of humans to perceive through touch. The findings will indicate whether or not tactile perception is optimal in the Bayesian sense. ******This work will have important theoretical and practical implications. The work will deepen basic scientific understanding of perceptual processing and may also inform the construction of next-generation tactile graphics displays, haptic virtual reality systems, and robotics. In order to design a successful virtual reality system, scientists and engineers must first understand how the end-user processes sensory input to arrive at a perceptual inference. Similarly, in order to create human-like robotics systems, scientists and engineers must first understand human perceptual processing. Our research will enable such understanding, thereby benefitting Canadian science and engineering.**