Under Pressure: Why the Brain Pays Attention

The idea that attention resolves competition between visual representations has a long history, much of it grounded in animal electrophysiology. Developing this account in humans is more difficult because neuroimaging measures distributed population activity rather than single-neuron responses. I’ll describe a set of experiments using computational modelling of EEG and concurrent EEG/MRI to characterise competition between multivariate object representations at the population level, asking how their representational geometry, strength, and directional interference shape attentional selection and suppression. Remarkably, this same competitive structure can be recovered from AI representations of visual objects, where it predicts the attentional responses observed in the human brain.