Résumé : There is ample neurobiological evidence that context-sensitive neocortical neurons use their apical inputs as context to amplify the transmission of coherent feedforward (FF) inputs. However, it has not been demonstrated until now how this known mechanism can provide useful neural computation. Here we show for the first time that the processing and learning capabilities of this form of neural information processing are well-matched to the abilities of mammalian neocortex. Specifically, we show that a network composed of such local processors restricts the transmission of conflicting information to higher levels and greatly reduces the amount of activity required to process large amounts of heterogeneous real-world data e.g., when processing audiovisual speech, these local processors use seen lip movements to selectively amplify FF transmission of the auditory information that those movements generate and vice versa. As this mechanism is shown to be far more effective and efficient than the best available forms of deep neural nets, it offers a step-change in understanding the brain's mysterious energy-saving mechanism and inspires advances in designing enhanced forms of biologically plausible machine learning algorithms.