Conference on Cognitive Computational Neuroscience

When presenting two targets (T1 and T2) embedded in a series of distractors, the ability to correctly report T2 decreases if T2 is shown between 200 and 500ms following T1, a phenomenon known as the attentional blink (AB). This decrease in performance is absent if T2 is shown immediately (or 100ms) after T1, and this is often referred to as lag-1 sparing. Here we used representational similarity analysis and deep convolutional neural networks trained on object categories to define the similarity between T1 and T2 and show that pairs of targets that share mid-level representations facilitate the conscious access of T2 at lag 1. This provides new insights on the mechanisms subtending the attentional blink (bottle-neck vs input-filter explanations) and the level of computation involved when processing the two targets.