- Annual Conference 2014
- About Us
- Apply to CoMPLEX
- For Students
- Students & Alumni
Modelling: Big Data and Society Conference
A new PhD student publication
Cognitive, Perceptual & Brain Sciences, UCL
Roberts Building 508 (<map>)
What do face aftereffects tell us about face representations?
After looking at a male face, people are more likely to categorise androgynous faces as female, and vice versa. It is widely thought that face aftereffects result from the adaptation of broadly-tuned channels signalling deviations from the average face, and so support prototype rather than exemplar models. This conclusion rests on the premises that face aftereffects are normalisation effects (in which the adapting stimulus becomes apparently more neutral), rather than local repulsion effects (in which the apparent difference between the adapting and nearby stimuli is exaggerated), and that the former pattern implicates pairs of broadly-tuned channels, while the latter implicates multiple narrowly-tuned channels. I will discuss two related lines of research questioning this conclusion.
First, most face aftereffect experiments have used binary classifications of single stimuli (e.g. “is this face male or female?”). This limits the number of test points at which one may measure changes in appearance, and makes it difficult to dissociate changes in criteria from changes in perception. When we used three-category rather than two-category classification tasks, facial gender and facial distortion aftereffects both produced (different) asymmetric patterns of category boundary shifts, more consistent with multichannel than opponent models. I am now exploring a two-alternative forced-choice double-adaptor task in which appearance at one adapted location is compared to appearance at a differently adapted location.
Second, normalisation may arise from the adaptation of multiple narrow channels. There are reports that tilted lines drift towards their nearest cardinal orientation with prolonged viewing, but several of these have not taken baseline measures, and could instead be caused by greater discrimination sensitivity at the cardinal axes. We are now developing a novel test of tilt normalisation, in which adaptors and tests of different spatial frequencies are compared after brief or long adaptation.
Page last modified on 26 jan 14 18:14