Motor, sensory, and cognitive learning require networks of neurons to generate new activity patterns. Because some behaviors are easier to learn than others1,2, we wondered if some neural activity patterns are easier to generate than others. We asked whether the existing network constrains the patterns that a subset of its neurons is capable of exhibiting, and if so, what principles define the constraint. We employed a closed-loop intracortical brain-computer interface (BCI) learning paradigm in which Rhesus monkeys controlled a computer cursor by modulating neural activity patterns in primary motor cortex. Using the BCI paradigm, we could specify and alter how neural activity mapped to cursor velocity. At the start of each session, we observed the characteristic activity patterns of the recorded neural population. These patterns comprise a low-dimensional space (termed the intrinsic manifold, or IM) within the high-dimensional neural firing rate space. They presumably reflect constraints imposed by the underlying neural circuitry. We found that the animals could readily learn to proficiently control the cursor using neural activity patterns that were within the IM. However, animals were less able to learn to proficiently control the cursor using activity patterns that were outside of the IM. This result suggests that the existing structure of a network can shape learning. On the timescale of hours, it appears to be difficult to learn to generate neural activity patterns that are not consistent with the existing network structure. These findings offer a network-level explanation for the observation that we are more readily able to learn new skills when they are related to the skills that we already possess3,4.

Efforts to study the neural correlates of learning are hampered by the size of the network in which learning occurs. To understand the importance of learning-related changes in a network of neurons, it is necessary to understand how the network acts as a whole to generate behavior. Here we introduce a paradigm in which the output of a cortical network can be perturbed directly and the neural basis of the compensatory changes studied in detail. Using a brain-computer interface, dozens of simultaneously recorded neurons in the motor cortex of awake, behaving monkeys are used to control the movement of a cursor in a three-dimensional virtual-reality environment. This device creates a precise, well-defined mapping between the firing of the recorded neurons and an expressed behavior (cursor movement). In a series of experiments, we force the animal to relearn the association between neural firing and cursor movement in a subset of neurons and assess how the network changes to compensate. We find that changes in neural activity reflect not only an alteration of behavioral strategy but also the relative contributions of individual neurons to the population error signal.brain-machine interface ͉ neural prosthetics ͉ perturbation ͉ population vector algorithm A wealth of evidence associates learning with changes in neural activity (for reviews, see refs. 1-4). For example, the tuning functions of neurons in the motor cortex can change when monkeys adapt to perturbations that interfere with the execution (5-7) or visual feedback (8 -10) of their movements. Although the observed changes in neural activity in these studies are closely associated with the behavioral manifestation of learning, it is difficult to interpret the behavioral significance of the neural changes because the precise relationship between neural activity and behavioral output is unknown. However, recent developments in brain-computer interface technology now make it possible to control the activity of a cursor in a three-dimensional (3D) virtual environment using the spiking activity of ensembles of simultaneously recorded motor cortical units (11)(12)(13)(14)(15)(16). This ''braincontrol'' paradigm is unique in that the behavior, cursor movement, is solely the result of neural activity in the population under study. Thus, any mismatches between desired cursor motion and decoded cursor motion can only be corrected by altering the activity of these recorded neurons.The brain-computer interface also allows for a unique kind of perturbation that targets selective subsets of neurons. By altering the way that the firing activity of a subset of neurons is decoded and mapped into cursor movement, it is possible to test whether the tuning functions of the perturbed subset selectively change to compensate for the global error signal, or whether the entire population changes together.To study the response to selective perturbations of this neural network, monkeys were trained to perform center-out movements to eight equally spaced targets in a 3D virtual environm...