The cholinergic system has been associated with learning, but also with cognitive decline in dementia and aging. Yet, to date no computational models have been put forward to explain how the cholinergic system contributes to both learning and cognitive decline. Here we introduce a model that combines a recently proposed biologically plausible model of credit assignment with a cholinergic adaptive module based on adaptive deep learning rules. Using a multi-class perceptual discrimination task we show that cholinergic adaptive learning leads to rapid learning when compared with non-adaptive models. Our results suggest that this is a consequence of the reorganizing effect that these forms of adaptive learning have, which encourages the network to develop distributed neuronal representations with mixed-selectivity. As a consequence, we show that the network becomes more robust to perturbations such as simulated cell death. Moreover, we demonstrate that in order to obtain such rapid and robust learning, it is not sufficient for cholinergic modulation to operate at a purely global level, suggesting an important function for local sources of acetylcholine in the cortex, such as ChAT-VIP interneurons. Overall, our work provides a novel theoretical framework for cellular-systems neuroscience with which to link cholinergic modulation of cortical learning to its function in health and disease.