The Weighting Strategy Of Multi-feature Stimuli Ensemble Representation

Ensemble representation refers to the cognitive process of summarizing similar(structured)stimuli in a highly compressed form.A relatively accurate ensemble representation overcomes the visual processing bottleneck and also assists human to obtain a rich visual experience.The classical proposal holds that the ensemble representation is independent of attention and individual representations.But,recent studies found the phenomenon of weighting and sampling in the ensemble representation,which challenged and questioned the classical proposal on the processing mechanism of ensemble representation.However,the role of individual stimuli on the weighting of the ensemble representation is not clear,and random weighting,equal weighting,and reliability weighting have different standpoints.Meanwhile,some scholars have found that the salience of individual stimuli has a significant impact on the weighting of the ensemble representation,that is,although the top-down attention is almost absent or weak,the bottom-up salience factors(such as the category boundary and the homogeneity)can bring about differences in the weighting of individual stimuli.In addition,previous studies on the weighting of the ensemble representation were few in number and mostly concentrated on the single-feature stimuli or the single feature in multi-feature stimuli.In light of this,this research intended to start from the influence of salient factors on the single-feature ensemble representation weighting strategy to investigate the interaction and priority of the single-feature weighting strategy in the multi-feature ensemble representation,further analyzing the weighting strategy of the multi-feature ensemble representation through its model exploration.This research helped to reveal the processing mechanism of the multi-feature ensemble representation and provided a new perspective and basis for the development of intelligent systems and other relevant systems.Considering the above purposes,this research was composed of four parts.Study 1 applied the single-feature stimuli,the selectivity ratio and sampling efficiency(as indicators),and a modified version of the centroid paradigm to manipulate the category boundary and homogeneity to examine the weighting strategies of length(Experiment 1),shape(Experiment 2)and color(Experiment3).Similarly,Study 2 applied the multi-feature stimuli to examine the interaction of single-feature weighting within the multi-feature ensemble representation under the contexts of length-color feature conjunction(Experiment 4)and color-shape feature conjunction(Experiment 5).Study 3 adopted the region segmentation paradigm to explore the processing priority of single-feature weighting under the contexts of lengthcolor feature conjunction(Experiment 6)and color-shape feature conjunction(Experiment 7).Study 4(Experiment 8)attempted to build a prediction model for the multi-feature ensemble representation through model exploration of the linear regression model and the neural network models(BP and LSTM),further analyzing and verifying the previous results.The major conclusions are listed as follows:(1)The feature-based attention and individual representations have important impact on the ensemble representation of single-feature stimuli,in which the category boundary and homogeneity play significant roles;(2)In the ensemble representation of multi-feature stimuli,there is a certain interaction between the weightings of single-feature stimuli;(3)In the ensemble representation of multi-feature stimuli,the weighting of singlefeature stimuli follows a certain processing priority.The processing priority of length is higher than that of color under the context of length-color feature conjunction,the processing priority of color is higher than that of shape under the context of color-shape feature conjunction;(4)The weighting of multi-feature ensemble representation is the result of singlefeature weightings and their interaction,which is consistent with the linear regression model and BP neural network;(5)The visual system may perform ensemble representation through sampling.Under the current context of presenting 6 stimuli,a relatively accurate ensemble representation can be achieved by sampling about 3 stimuli,which conforms to the assumption of √n(n is the number of stimuli).This research demonstrates that the visual system performs differential weighting in the ensemble representation based on individual representations.For the ensemble representation of multi-feature stimuli,the weightings of different single-feature stimuli influence each other and follow a certain priority.The ensemble representation of multi-feature stimuli is similar to a self-learning/adaptive process.