New computational techniques [1, 2, 3, 4, 5, 6, 7] enable neural networks to be trained on tasks similar to those used in experiments with behaving animals [8, 9, 10, 11, 12, 13, 14, 15]. Before these techniques became available, a researcher would hypothesize what computations the network should perform to execute the task, and build a network architecture capable of carrying them out. Then, numerical simulations of the model or mean field approximations allowed verifying whether the proposed network model performed the task as desired. This is unsatisfactory, as it does not allow identifying how a neural network could solve these tasks; the models thus constructed only reflect the researcher's intuitions about how the tasks could be performed. In contrast, trained networks provide us with a valuable tool to investigate mechanisms that networks could use to perform the tasks [16, 17, 18, 19, 6, 7, 20, 13, 21, 22, 23].

Trained recurrent networks are powerful tools for modeling dynamic neural computations. We present a target-based method for modifying the full connectivity matrix of a recurrent network to train it to perform tasks involving temporally complex input/output transformations. The method introduces a second network during training to provide suitable "target" dynamics useful for performing the task. Because it exploits the full recurrent connectivity, the method produces networks that perform tasks with fewer neurons and greater noise robustness than traditional least-squares (FORCE) approaches. In addition, we show how introducing additional input signals into the target-generating network, which act as task hints, greatly extends the range of tasks that can be learned and provides control over the complexity and nature of the dynamics of the trained, task-performing network.