Recurrent neural networks that learn multi-step visual routines with reinforcement learning

We report that networks learn elemental operations, such as contour grouping and visual search, and execute sequences of operat ions, solely based on the characteristics of the visual stimuli and the reward structure of a task. After training was completed, the activity of the units of the neural network elicited by behaviorally relevant image items was stronger than that elicited by irrelevant ones, just as has been observe d in the visual cortex of monkeys solving the same tasks. Relevant information that needed to be exchanged between subroutines was maintained as a focus of enhanced activity and passed on to the subsequent subroutines. Our results demonstrate how a biologically plausible learning rule can train a re current neural network on multistep visual tasks.
Source: PLoS Computational Biology - Category: Biology Authors: Source Type: research