 Deep neural networks optimized for visual tasks can learn representations that align layer depth with the hierarchy of visual areas in the primate brain. However, these representations do not necessarily need to be hierarchical, as demonstrated by a recent study which showed that a single-layer network could also accurately predict brain activity in V1, V4. The study suggests that DNNs may have different architectures, ranging from strict serial hierarchies to multiple independent branches. This article was authored by Gislin S. T. Eve, Emily J. Allen, Ihan Wu, and others.