Visual system articles from across Nature Portfolio

Determining what infants recognize in the things they see has been difficult. Using brain imaging and deep neural networks, we found that in 2-month-old infants the visual system represented not only how things looked but also what category they belonged to. This showed that features sufficient for object recognition are already important in early life.

Using infant fMRI, the authors show that, by 2 months of age, representations in high-level visual cortex encode visual categories that align with deep neural networks, and lateral object-selective regions are later to develop.

How the human brain transforms visual input into meaningful object representations remains unclear. Here, the authors show that dense feature-based coding in the ventral temporal cortex is transformed into sparse, conceptual representations in the medial temporal lobe, revealing neural computational mechanisms of object recognition.

Objects in natural scenes are often partially occluded. Here, the authors show that recurrent processing can use knowledge of the occluder to explain away missing features, improving recognition of occluded objects in models and humans.

It is currently unknown how uniform is signalling at the first synapse of vision in vivo. Here, the authors show neighbouring PR1 (red) cones differ in contrast sensitivity and timing due to locally stochastic horizontal-cell feedback, thereby extending total dynamic range.

Determining what infants recognize in the things they see has been difficult. Using brain imaging and deep neural networks, we found that in 2-month-old infants the visual system represented not only how things looked but also what category they belonged to. This showed that features sufficient for object recognition are already important in early life.

Functional connectomics is a new approach, involving calcium imaging of neuronal activity followed by correlated electron microscopy connectomics of the same neurons, that relates connections made by neurons to their in vivo function. I believe that this combined approach for studying structure and function will continue with ever-larger networks, including entire nervous systems.

Even before training, convolutional neural networks may reflect the brain’s visual processing principles. A study now shows how structure alone can help to explain the alignment between brains and models.

In this Journal Club, Benjamin Cowley discusses a 2009 tour-de-force that provided a recipe for constructing closed-loop algorithms by letting predictive models speak for themselves.

To perceive events as simultaneous despite differences in how sensory signals are generated and transmitted, the brain must preserve temporal coherence. We found that retinal ganglion cells adjust the speed with which they transmit neural signals, revealing a mechanism in the human retina to keep visual perception precisely timed.