A Developmental Network Model of Conscious Learning in Biological Brains

Abstract We have seen many advances in computational neuroscience. However, what is still missing is a holistic model that is broad and deep. Being broad, it approximates brain development across a broad range of species (e.g., from fruit flies to humans) without given any tasks cross lifetime. Being deep, it specifies an algorithm so that the model can be verified across a deep and fluid hierarchy of scales without symbolic functional modules, from neurotransmitters, to cells, to wiring, to brain patterning, to behaviors, to intelligence, to consciousness. Fluid means task-nonspecific and for any consciousnesses. This report proposes such a model called Developmental Network 3 (DN-3). A major extension from the predecessor DN-2 to DN-3 is that the new model starts from a single cell zygote so that developmental algorithm develops sensors, a large brain and motors in a coarse-to-fine way, so that wiring and patterning are fully automatic from conception to death. The DNs have been supported by experimental simulations with real sensory data for vision, audition, natural languages, and planning, but this theoretical paper does not present new experiments. This neuromorphic model is informed by, and will be further refined by, biological studies.