Abstract (submitted version)
Generalization of learning refers to the phenomenon in which knowledge learned in one context enhances performance in another context. Although the learning environment can never be precisely the same in the real world, animals including humans demonstrate excellent flexibility to adapt their learned skills in a new environment. Despite the universal occurrence of generalization phenomena in daily life, there is much lack of understanding about how the brain generalizes the skills and knowledge into different environments. In particular, most of the previous studies have only focused on identifying the cerebral networks used during the generalization stage, but failed to determine the elements during the preceding learning stage that could have enabled generalization. Thus, the aim of this study was to enhance understanding of the neural mechanisms that enable generalization, particularly the generalization of motor learning. In this study, we designed a new experimental paradigm called 'mirror-erasing generalization task.' The subjects erased (1) a simple shape (square) for the training session, and (2) a complex shape (cursive alphabet letter y) for the generalization session, which took place both before and after the training session, in an MRI scanner. We found that the subjects successfully generalized their motor skills (p<0.0001) acquired during square-erasing to the letter-erasing context. However, counterintuitively, skill improvement during training did not correlate with the generalization of motor skills (p~0.1). This result implicates that the dynamics underlying generalization is possibly nonlinear, and that performance enhancement in one specific context is not a reliable measure to estimate the generalization performance in another. Then, we computationally modeled the neuronal circuitry responsible for motor learning generalization and used the fMRI machine to construct a functional network model (using frequency between 0.049Hz and 0.09Hz). More interestingly, we found that the betweenness centrality of pars opercularis of left inferior frontal gyrus (IFG) had a significant correlation with the generalization performance (R~0.8, p<0.05, FWE corrected), which was the only measure before generalization session that correlated with generalization performance. We should note that the human pars opercularis of the left IFG has been considered as part of a mirror neuron system, which is currently hypothesized to be facilitating motor abstraction. This finding suggests that the IFG-mediated abstraction of new motor skill acquired during training may be the key to generalization of the learned skills in a different context. This study potentially provides an evidence for the contemporary view that abstraction plays an essential role in generalization. Furthermore, we suggest that the centrality of par opercularis of the left IFG is possibly used to make predictions about future generalization performance, which opens new possibilities in motor rehabilitation. This study suggests that measuring brain functional networks of the patients undergoing rehabilitation programs potentially predict how much their motor function would be improved in real life.
Acknowledgement
This study was conducted as part of Global Singularity Research Program for 2020 financially supported by KAIST.