Lateral inhibition is when a groups of nodes at the same layer in a brain or artficial neural network have interconnections that inhibit one another. This is may encode aspects where there are different interpretations of a visual scene or data that are effectively fighting against one other – the blob to the right of the scene is a pussy cat or a seal, but can't be both at the same time. Lateral inhibition may itself have a direction, but more often involves mutual inhibition. Where a neural network is used for classification there is often a winner takes all or top K selection at the final layer, which is often seen as separate to the network itself, but could be viewed as a form lateral inhibition. Memory layers apply the same mechanism at an internal leyer to represent key-value memories.
Used in Chap. 6: page 92