Selective distractor representations resolve multidimensional interference

How can humans manage multiple sources of information competing for attention? To approach this question, we adopted a multi-dimensional task-set interference paradigm that requires individuals to handle distractions from three independent dimensions. Behavioral results suggest that people track prior interference from each dimension to selectively modulate their attentional gain. Testing the mechanism of this adaptation at the neural level requires measuring multi-dimensional task representations. To achieve this, we applied representational similarity analyses and encoding models to human EEG and investigated how the history of interference simultaneously affected the time-resolved representations of target and distractor dimensions. EEG analyses revealed that target and distractor features are initially encoded in parallel, but distractors are rapidly suppressed around 250 ms, with suppression scaling with prior interference. Next, we introduced a task-specific proportion-congruency manipulation to study how learning the control demands of each task dimension shapes the proactive handling of multidimensional distractors. Proactive task-level control enhanced the efficiency of the same reactive suppression mechanism observed for trial-to-trial adaptation, without producing sustained preparatory changes. Finally, behavioral and neural effects converged to show that the strength of dimension-specific interference and adaptation scales with the speed of visual integration for each dimension, as captured by a connectionist model with temporal integration. Altogether, these results suggest that proactive control mechanisms operate on the speed and efficiency of a reactive suppression mechanism that constrains selective distractor representations from biasing the response process. More broadly, they show that multidimensional attentional control relies on selectively suppressing distractor representations after they are encoded, revealing a dynamic, dimension-specific mechanism that extends biased-competition and conflict-monitoring theories.