Abstract
Models of visual working memory (VWM) have recently been assessed using cued feature report, predicting performance for clearly visible stimuli by estimating the probability of being in VWM and the variability of feature encoding (Zhang & Luck, 2008; Bays, et al., 2009). However, VWM presumably must operate on low contrast and noisy stimuli, not just on highly visible ones. We report an integrated multi-alternative template model of VWM with stimulus variations of contrast and external noise that provide demanding new tests. 900 ms after display, the observers were cued to report the orientation of one of up to four briefly presented Gabors from a selection of 20 orientations (every 9°). We measured display size effects (1, 2, 4) with and without external noise for Gabors at eight contrast levels. Memory report was systematically affected by display size, Gabor contrast, and external noise, varying from poor to excellent. Yet, there is a surprising similarity of the spread of the reports around the correct orientation for conditions at many levels of performance. A multi-alternative perceptual template model provides a full account of the data. Twenty templates respond with gains depending on the match between each template and the cued target stimulus. Additive and multiplicative noises are added to the outputs of the templates. A max rule chooses the response. Focal attention in smaller displays enhances the representations by reducing internal additive noise and improved filtering of external noise, so estimates of total internal noise increase with display size (Wilken & Ma, 2004). However, the spread of the distribution of the reported orientations does not vary with display size – it is determined by the tuning of the competing templates. The multi-alternative template model provides a parameter-efficient and counter-intuitive integrated account of VWM and attention over many display conditions, challenging existing VWM model.
Meeting abstract presented at VSS 2012