Selective distractor representations resolve multidimensional interference
Navigating competing attentional demands is a core cognitive function, yet how the brain tunes control in multidimensional environments remains poorly understood. Here, we used a multidimensional task-set interference paradigm in which participants attended to one of four stimulus dimensions while three others acted as distractors, combining multivariate decoding, representational similarity analysis, and encoding models applied to human EEG. Targets and distractors were initially encoded in parallel, but distractor representations were rapidly suppressed [~]250 ms after stimulus onset, with suppression scaling with each distractors own conflict history. Neither trial-to-trial adaptation nor block-level learning produced anticipatory changes in task-relevant representations. Instead, proactive control modulated the speed and efficiency of stimulus-triggered suppression. Encoding models further revealed that conflict is represented in orthogonal, dimension-specific subspaces that eventually collapse onto a shared conflict signal. These results show that multidimensional attentional control operates through selective, reactive suppression of distractor representations, guided by a structured multivariate conflict signal.