Westerberg, J. A.; Xiong, Y. S.; Nejat, H.; Sennesh, E.; Durand, S.; Cabasco, H.; Belski, H.; Gillis, R.; Loeffler, H.; Bawany, A.; Peene, C. R.; Han, W.; Nguyen, K.; Ha, V.; Johnson, T.; Grasso, C.; Hardcastle, B.; Young, A.; Swapp, J.; Ouellete, B.; Caldejon, S.; Williford, A.; Groblewski, P. A.; Olsen, S. R.; Kiselycznyk, C.; Lecoq, J. A.; Maier, A.; Bastos, A. M.

Predictive coding (PC) is a popular framework to explain cortical responses. PC states that the brain computes internal models of expected events and responds robustly to unexpected stimuli with prediction errors that feed forward. This can be studied using oddball tasks. A repetitive sequence interrupted by a novel stimulus is a \"local\" oddball. \"Global\" oddballs defy expectations while repeating the local context. This dissociates expectation-driven neuronal prediction error responses vs. merely stimulus-history-driven release from adaptation. We recorded neuron spiking activity across the visual hierarchy in mice and monkeys viewing these oddballs. Local oddball responses largely followed PC hypotheses. They were robust, emerged early in layers 2/3, and fed forward. Global oddball responses challenged PC. They were weak, absent in most visual areas, more robust in prefrontal cortex, and emerged in non-granular layers. Contrary to the most popular PC models that emphasize prediction error computations occurring in early sensory areas and feeding forward, this implies that genuine predictive coding computations take place in more cognitive, higher-order areas.