I did my secondment at the Cognition and Development Department, Institute of Neurosciences, Univertat de Barcelona in Joan Lopez-Moliner’s team from January 15th to March 2nd 2018.
Internal models, systems which mimic the behavior of natural computational processes in the brain, has been studied for many years by human motor control theories. There are two types of internal model: (i) forward models, which try to estimate the sensory consequences of our executed motor command; and (ii) inverse models, which choose suitable sets of motor command to achieve to a desired state of our body. While implicitly our internal models update and get adapted within different contexts, explicitly we might not be aware of the state of surrounding environment. The bidirectional impact of implicit and explicit processes in the brain and their influence on our net motor command has been started to be studied only quite recently. Mazzoni and Krakauer  asked a group of participants to make reaching movements while they perturbed their feedback of the hand by applying a 45 degree of visual perturbation. The first group of participants gradually adapted and learned how to make movement in response to the perturbation. In contrast, the second group of participants after making two large errors, were told about the visual perturbation that was applied. Although participants could reach to the target right after the strategy has been told, they couldn’t keep up their performance and the end-point error increased for the rest of trials. This wa observed since, while explicitly aware about the perturbation and the way to counter it, their sensorimotor system implicitly adapts.
In order to better understand the interaction between these two processes (based on a data which has been collected before), we tried to model our behavioral data set by using the Kalman filter approach. The generalization of motor learning depends on both prediction errors and the history of prior actions. The Kalman framework takes care of both prior and current observations by weighting the prediction error based on the ratio between uncertainties of our prediction and our incoming sensory information.
The task has been designed such that the participants are trained to compensate for a visual perturbation (30 degree) explicitly using a cue. After a catch trial, a cue appears, and the participants are asked to counter the visual perturbation by pointing to the adjacent target for three trials. Then, the visual perturbation disappears, which the participants are informed of. These sets of trials have been repeated and the perpendicular deviation has been computed as the error on a given trial.
Picture: Different types of trials in the task
Picture: PD computed for different types of trials
By applying the Kalman filter approach, we could dissociate two hidden processes which are responsible for generating the observed patterns. The first process has a smaller retention rate (forgetting factor) but the learning rate (Kalman gain) of the second process is higher. These hidden states are consistent with a study by Smith and colleagues , in which they show that saving or spontaneous recovery can be explained by considering two, fast and slow changing, hidden processes contributing to the net motor output.
Our results are consistent with the idea that potential candidates for these two processes are implicit and explicit adaptation processes of the brain. The explicit process forgets faster but also learns faster, and in contrast the implicit process remembers for longer period of time but learns much more slowly.
References:. Mazzoni P, Krakauer JW (2006) An implicit plan overrides an explicit strategy during viusomotor adaptation. Journal of Neuroscience 26(14):3642–3645. . Smith MA, Ghazizadeh A, Shadmehr R (2006) Interacting adaptive processes with different timescales underlie short-term motor learning. PLoS Biol 4(6): e179. DOI:10.1371/journal.pbio.0040179.