Project A3. Learning and adaptation of spatial attention templates in visual search

在日常场景中，搜索的目标不会孤立地出现，而是嵌入在非目标或干扰项的配置中。如果目标相对于干扰物的位置是不变的，则这种空间偶然性被隐含地学习并且开始引导视觉扫描（上下文提示效应; Chun & Jiang，1998;盖耶等人，2010年b）。在预测编码模型方面，内隐神经学习可以被概念化为涉及输入和原因之间的关联的逐渐获得（Friston，2010; Conci等人，2012年）。那就是：搜索引导预测（关于目标的位置）从上下文存储器自上而下地传播，以影响自下而上的搜索过程。然而，上下文预测的有效性在很大程度上取决于自下而上的感知输入和上下文记忆之间的一致性：在图形学习之后，将目标重新定位到未改变的干扰物上下文中的意外位置完全消除了上下文提示，并且随着对改变的显示的暴露增加，从不变上下文获得的增益恢复得非常缓慢（Zellin等人。2014）。本研究将采用行为学和神经科学的方法来研究视觉搜索中语境学习的自上而下和自下而上的平衡。

In everyday scenes, searched-for targets do not appear in isolation, but are embedded within configurations of non-target or distractor items. If the position of the target relative to the distractors is invariant, such spatial contingencies are implicitly learned and come to guide visual scanning (contextual-cueing effect; Chun & Jiang, 1998; Geyer et al., 2010b). In terms of predictive coding models, implicit configural learning can be conceptualized as involving the gradual acquisition of associations between input and cause (Friston, 2010; Conci et al., 2012). That is: search-guiding predictions (about the location of a target) are propagated top-down from contextual memory to influence bottom-up search processes. However, the effectiveness of contextual predictions depends heavily on the consistency between bottom-up perceptual input and context memory: following configural learning, re-locating targets to an unexpected location within an unchanged distractor context completely abolishes contextual cueing, and the gains deriving from the invariant context recover only very slowly with increasing exposure to the changed displays (Zellin et al. 2014). The present proposal will employ behavioral and neuroscience methods to investigate the top-down-bottom-up balance of contextual learning in visual search.