Search Papers | Poster Sessions | All Posters
Poster C31 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
The Learning Hypothesis on Spatial Receptive Field Remapping
Henry Kuo1, Paul Masset2, Blake Bordelon1, Cengiz Pehlevan1 (); 1Harvard University, 2McGill University
Grid cells in the medial entorhinal cortex (MEC) and place cells in the hippocampus fire only at specific locations in the environment. Their firing fields are known to change to spatial and nonspatial alterations of the environment. A recent study showed that their receptive fields move toward rewards after rats learn the locations of the rewards. We hypothesize that this change in receptive fields could help animals learn better. To verify this hypothesis, we use receptive fields as features in a simulated agent performing Temporal Difference learning. First, we observed that an evolutionary algorithm would move mathematical place cells closer to rewards to speed up learning. Second, using rat MEC neural data from the study, we found that the simulated agent can learn better with the post-learning receptive fields when evaluating post-learning behavior. These results show that distortions in spatial receptive fields could be a feature of the brain to help animals perform learning tasks.
Keywords: grid cells spatial field remapping temporal difference learning reinforcement learning