The study investigates the context-dependent computation in prefrontal cortex (PFC) by examining the activity of neurons in macaque monkeys performing a task that involves selecting and integrating noisy sensory inputs. The researchers found that the complex responses of individual neurons can be understood within a dynamical process at the population level. They developed a trained recurrent neural network that reproduces the observed population dynamics, suggesting a novel mechanism for selecting and integrating task-relevant inputs. This mechanism involves two key features: an approximate line attractor and a 'selection vector,' which are defined at the population level. The line attractor represents the relevant input, while the selection vector orthogonalizes the irrelevant input, allowing for context-dependent integration. This framework provides a general solution to context-dependent computations and highlights the role of PFC in flexible behavior.The study investigates the context-dependent computation in prefrontal cortex (PFC) by examining the activity of neurons in macaque monkeys performing a task that involves selecting and integrating noisy sensory inputs. The researchers found that the complex responses of individual neurons can be understood within a dynamical process at the population level. They developed a trained recurrent neural network that reproduces the observed population dynamics, suggesting a novel mechanism for selecting and integrating task-relevant inputs. This mechanism involves two key features: an approximate line attractor and a 'selection vector,' which are defined at the population level. The line attractor represents the relevant input, while the selection vector orthogonalizes the irrelevant input, allowing for context-dependent integration. This framework provides a general solution to context-dependent computations and highlights the role of PFC in flexible behavior.