. A computational model of muscle recruitment for wrist movements.

No abstract

Change in behavior and neural activity in skill acquisition suggests that control is transferred from cortical planning areas (e.g., the prefrontal cortex, PFC) to the basal ganglia (BG). Planning has large computational and representational requirements but requires little experience with a task. The BG are thought to employ a simpler control scheme and reinforcement learning; these mechanisms rely on extensive experience. Many theoretical accounts of behavior in the face of uncertainty invoke planning mechanisms that explicitly take uncertainty into account. We suggest that the simpler mechanisms of the BG can also contribute to the development of behavior under such conditions. We focus on learning under conditions in which sensory information takes time to resolve, e.g., when a poorly perceived goal stimulus takes non-negligible time to identify. It may be advantageous to begin acting quickly under uncertainty -possibly via decisions that are suboptimal for the actual goal -rather than to wait for sensory information to fully resolve.We present a model of skill acquisition in which control is transferred, with experience, from a planning controller (denoted A), corresponding to the PFC, to a simpler controller (B), corresponding to the BG. We apply our model to a task in which a learning agent must execute a series of actions to achieve a goal (selected randomly at each trial from a small set). Over the course of a trial, the agent's goal representation evolves from representing all possible goals to only the selected goal. A is restricted to select movements only when goal representation is fully resolved. Model behavior is similar to that observed in humans accomplishing similar tasks. Thus, B can by itself account for the development of behavior under an evolving sensory representation, suggesting that the BG can contribute to learning and control under conditions of uncertainty.

. Cortical involvement in the recruitment of wrist muscles. J Neurophysiol 91: 2445-2456, 2004. First published January 28, 2004 10.1152/jn.00879.2003. In executing a voluntary movement, one is faced with the problem of translating a specification of the movement in task space (e.g., a visual goal) into a muscle-recruitment pattern. Among many brain regions, the primary motor cortex (MI) plays a prominent role in the specification of movements. In what coordinate frame MI represents movement has been a topic of considerable debate. In a two-dimensional wrist step-tracking experiment, Kakei et al. described some MI cells as encoding movement in a muscle-coordinate frame and other cells as encoding movement in an extrinsic-coordinate frame. This result was interpreted as evidence for a cascade of transformations within MI from an extrinsic representation of movement to a muscle-like representation. However, we present a model that demonstrates that, given a realistic extrinsic-like representation of movement, a simple linear network is capable of representing the transformation from an extrinsic space to the muscle-recruitment patterns implementing the movements on which Kakei et al. focused. This suggests that cells exhibiting extrinsic-like qualities can be involved in the direct recruitment of spinal motor neurons. These results call into question models that presume a serial cascade of transformations terminating with MI pyramidal tract neurons that vary their activation exclusively with muscle activity. Further analysis of the model shows that the correlation between the activity of an MI neuron and a muscle does not predict the strength of the connection between the MI neuron and muscle. This result cautions against the use of correlation methods as a measure of cellular connectivity.

Animals, interacting with the environment, learn and exploit the consequences of their movements. Fundamental to this is the pairing of salient sensory input with recent motor output to form an action-outcome pair linking a performed movement with its outcome. Short-latency dopamine (DA) signalling in the basal ganglia has been proposed to support this crucial task. For visual stimuli, this DA signalling is triggered at short latency by input from the superior colliculus (SC). While some aspects of the visual signal (e.g. luminance), are relayed directly to the SC via the retinotectal projection, other information unavailable to this subcortical pathway must take a more circuitous route to the SC, first submitting to early visual processing in cortex. By comparing action-outcome pairing when the visual stimulus denoting success was immediately available to the SC, via the retinotectal pathway, against that when cortical processing of the signal was required, the impact this additional sensory processing has on action-outcome learning can be established. We found that action acquisition was significantly impaired when the action was reinforced by a stimulus ineligible for the retinotectal pathway. Furthermore, we found that when the stimulus was eligible for the retinotectal pathway but evoked an increased latency, action acquisition was not impaired. These results suggest that the afferent sensory pathway via the SC is certainly primary and possibly instrumental to the DA neurons' role in the discovery of novel actions and that the differences found are not due to simple sensory latency.

No abstract

Many tasks, such as typing a password, are decomposed into a sequence of subtasks that can be accomplished in many ways. Behavior that accomplishes subtasks in ways that are influenced by the overall task is often described as "skilled" and exhibits coarticulation. Many accounts of coarticulation use search methods that are informed by representations of objectives that define skilled. While they aid in describing the strategies the nervous system may follow, they are computationally complex and may be difficult to attribute to brain structures. Here, the authors present a biologically- inspired account whereby skilled behavior is developed through 2 simple processes: (a) a corrective process that ensures that each subtask is accomplished, but does not do so skillfully and (b) a reinforcement learning process that finds better movements using trial and error search that is not informed by representations of any objectives. We implement our account as a computational model controlling a simulated two-armed kinematic "robot" that must hit a sequence of goals with its hands. Behavior displays coarticulation in terms of which hand was chosen, how the corresponding arm was used, and how the other arm was used, suggesting that the account can participate in the development of skilled behavior.

No abstract