People can extract ensemble representations from different types of visual stimuli, for example, average location, size or direction from sets of dots, circles and gabors, as well as average emotion, identity or attractiveness from sets of faces. However, ensemble representations of facial emotion, identity and attractiveness were observed to be different from that of the physical attributes of simple geometry stimuli in many aspects. Firstly, the relationship between the member representation and the ensemble representation seemed to vary dependent on the stimuli and the attributes. For simple stimuli and physical attributes, the feature of the set could be estimated accurately while the features of the members might have been lost or discarded. But for faces and facial attributes, the features of the members seemed retained and could support successful identification. Secondly, the effects of set size and duration were reported to differ between different stimulus ensembles and different attributes. For simple stimuli and physical attributes, the process of ensemble representation remained relatively unchanged when the set size expanded or the duration shortened. On the other hand, for faces and facial attributes, the same process seemed to be partly affected by the change of set size and duration, showing some sensitivity to the amount of information presented in a given period of time. Thirdly, the effect of member variability upon ensemble representation was also recorded to differ. For simple stimuli and physical attributes, the degree of variation among members could influence the accuracy of estimating the mean of the set. But for faces and facial attributes, the

accuracy of the estimation remained the same despite increase of variation. These observed differences suggest that the underlying cognitive mechanism for ensemble representation may vary from stimulus to stimulus, from attribute to attribute. Parallel processing and texture analysis may be dominant in the ensemble representation of simple stimuli, especially when estimating the set mean of physical attributes like location, size or direction. This could be the reason why an ‘advantage of the whole’ was reported and the mean estimation would prevail even when the identification of members failed. Meanwhile, sequential processing and facial holistic analysis may be essential in the ensemble representation of faces. Because every single face was processed as an independent and holistic subject, information of member faces was retained during ensemble representation and the identification of member stimuli could thus be accomplished. In light of these differences in both observations and mechanisms, future research should try to separate the perception of individual set members and the representation of the set so as to directly compare the ensemble processing of various types of stimuli with different attributes. Furthermore, it would be worthwhile to explore ensemble representations of social dimensions inferred by faces, like credibility, social openness and so on.

从点、圆、光栅集合的中心位置、平均大小、平均朝向, 到面孔集合的平均情绪、身份和吸引力, 人们可以通过集群表征的方式加工各种刺激集合. 但简单几何刺激在物理属性上的集群表征, 与面孔刺激在情绪、身份和吸引力等复杂属性上的集群表征存在差异, 主要表现在成员个体表征与集群表征的关系, 集合容量和呈现时间对集群表征的影响, 以及成员变异性对集群表征的影响等方面. 这些外在表现上的差异可能反映了内在机制上的差异: 简单刺激在物理属性上的集群表征可能更凸显平行加工的特点, 基于纹理性加工机制; 而面孔刺激在复杂属性上的集群表征可能更凸显序列加工的特点, 受到面孔整体性加工机制的影响. 未来研究可能需要分离刺激加工和集群表征两个过程, 关注集群表征在不同刺激的不同属性间的可比性; 同时探讨面孔集合社会属性的集群表征,在更为复杂的维度中考察集群表征现象.