@article{202791,
  keywords = {Motor learning, context learning, meta-learning, reinforcement learning, navigation},
  author = {Carlos Vel{\'a}zquez-Vargas and Isaac Christian and Jordan A. Taylor and Sreejan Kumar},
  title = {Learning to abstract visuomotor mappings using meta-reinforcement learning},
  abstract = {<p>We investigated the human capacity to acquire multiple visuomotor mappings for<br>de novo skills. Using a grid navigation paradigm, we tested whether contextual<br>cues implemented as different {\textquotedblright}grid worlds{\textquotedblright}, allow participants to learn two distinct<br>key-mappings more efficiently. Our results indicate that when contextual<br>information is provided, task performance is significantly better. The same held<br>true for meta-reinforcement learning agents that differed in whether or not they<br>receive contextual information when performing the task. We evaluated their accuracy<br>in predicting human performance in the task and analyzed their internal<br>representations. The results indicate that contextual cues allow the formation of<br>separate representations in space and time when using different visuomotor mappings,<br>whereas the absence of them favors sharing one representation. While both<br>strategies can allow learning of multiple visuomotor mappings, we showed contextual<br>cues provide a computational advantage in terms of how many mappings<br>can be learned.</p>},
  year = {2024},
  journal = {ICLR 2024 Workshop on Representational Alignment (Re-Align)},
  volume = {46},
  pages = {1-11},
  month = {07/2024},
  url = {https://escholarship.org/uc/item/4jj4q4df},
}