Tanaka Shingo, Oguchi Mineki, Sakagami Masamichi
Brain Science institute, Tamagawa University.
Brain Nerve. 2016 Nov;68(11):1263-1270. doi: 10.11477/mf.1416200590.
To behave appropriately in a complex and uncertain world, the brain makes use of several distinct learning systems. One such system is called the "model-free process", via which conditioning allows the association between a stimulus or response and a given reward to be learned. Another system is called the "model-based process". Via this process, the state transition between a stimulus and a response is learned so that the brain is able to plan actions prior to their execution. Several studies have tried to relate the difference between model-based and model-free processes to the difference in functions of the lateral prefrontal cortex (LPFC) and the striatum. Here, we describe a series of studies that demonstrate the ability of LPFC neurons to categorize visual stimuli by their associated behavioral responses and to generate abstract information. If LPFC neurons utilize abstract code to associate a stimulus with a reward, they should be able to infer similar relationships between other stimuli of the same category and their rewards without direct experience of these stimulus-reward contingencies. We propose that this ability of LPFC neurons to utilize abstract information can contribute to the model-based learning process.
为了在复杂且不确定的世界中做出恰当行为，大脑会利用多种不同的学习系统。其中一种系统被称为“无模型过程”，通过这种过程，条件作用使得刺激或反应与特定奖励之间的关联得以学习。另一种系统被称为“基于模型的过程”。通过这个过程，刺激与反应之间的状态转换得以学习，从而使大脑能够在行动执行之前进行规划。多项研究试图将基于模型和无模型过程之间的差异与外侧前额叶皮层（LPFC）和纹状体功能的差异联系起来。在此，我们描述了一系列研究，这些研究证明了LPFC神经元能够根据相关行为反应对视觉刺激进行分类并生成抽象信息。如果LPFC神经元利用抽象编码将刺激与奖励联系起来，那么它们应该能够在没有直接经历这些刺激 - 奖励关联的情况下，推断同一类别的其他刺激与其奖励之间的相似关系。我们提出，LPFC神经元利用抽象信息的这种能力有助于基于模型的学习过程。
