Neural Circuits, Vision, and Computation

## Computation over time Visual scenes change constantly. Retinal circuits do not process each moment independently; they use recent history. This is **temporal processing**. ## Adaptation is not fatigue When the **statistics** of the input change — for example, contrast rises from low to high — many retinal neurons change their sensitivity within seconds. This is **adaptation**. It is tempting to think of it as neurons getting "tired," but that is not the right picture. Adaptation adjusts the cell's operating range to match the current input statistics so that responses remain informative. After the change, the cell often still responds reliably; it has rescaled, not failed. Some cells do the opposite: after strong stimulation, they temporarily become *more* sensitive. This **sensitization** can be interpreted as preparing the circuit for a likely future signal. Adaptation and sensitization can coexist in the same retina, produced by different circuit components, often involving **inhibitory** amacrine-cell pathways. ## Prediction without foresight Natural scenes have **statistical regularities**: objects move smoothly, and textures persist. Retinal circuits can exploit these regularities. For instance, responses to a smoothly moving object can be shifted so that the population signal lines up closer to the object's current position, partly compensating for processing delays. This is sometimes called prediction, but it does **not** mean the retina consciously "sees the future." It means the circuit's transformation is shaped by regularities in the input, so its output carries information about what is likely to happen next. ## Object motion versus global motion When your eyes move, the whole image shifts — **global motion**. When a predator moves, only part of the image moves differently — **object motion**. Some ganglion cells respond to local motion only when it differs from the background. Inhibitory circuitry helps by suppressing responses when the center and the surround move together. This is a clear example of a computation (detect differential motion) with an identifiable circuit mechanism (wide-field inhibition). ## Further reading - [Baccus Lab research overview](https://baccuslab.github.io/research/) - [Baccus Lab publications](https://baccuslab.github.io/publications/)