Top-down feedback matters: Functional impact of brainlike connectivity motifs on audiovisual integration

Artificial neural networks (ANNs) can generate useful hypotheses about neural computation, but many features of the brain are not captured by standard ANNs. Top-down feedback is a particularly notable missing feature. Its role in the brain is often debated, and it’s unclear whether top-down feedback would improve an ANN’s ability to model the brain. Here we develop a deep neural network model that captures the core functional properties of top-down feedback in the neocortex. This feedback allows identically connected recurrent models to have different processing hierarchies based on the direction of feedforward and feedback connectivity. We then explored the functional impact of different hierarchies on audiovisual categorization tasks. We find that certain hierarchies, such as the one seen in the human brain, impart ANN models with a light visual bias similar to that seen in humans while maintaining excellent performance on all audio-visual tasks. The results further suggest that different configurations of top-down feedback make otherwise identically connected models functionally distinct from each other and from traditional feedforward-only recurrent models. Altogether our findings demonstrate that top-down feedback is a relevant feature of biological brains that improves the explanatory power of ANN models in computational neuroscience.