Hippocampal remapping as hidden state inference
- Center for Brains Minds and Machines, Harvard University, United States
- Picower Institute for Learning and Memory and Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, United States
- Department of Psychology, Harvard University, United States
Figures
Figure 1
The hidden state inference framework.
(A) Schematic of hidden state inference. We impute an internal generative model to the animal, according to which observations are generated by a small number of hidden states. States are sampled …
Figure 2
Hidden state inference is informed by cue constellations.
Observations are generated from a distribution with four features, each drawn from a Gaussian with mean 0 and standard deviation of 0.2.We train the model with 20 observations drawn from that …
Figure 3
Learning to distinguish.
(A) Adapted from Lever et al., 2002, who compared place cell representations between alternating presentations of square and circle boxes. Field Divergence is expressed in percent and represents the …
Figure 4
Map stabilization requires certainty about distributional statistics.
(A) Data from Law et al., 2016, showing the spatial correlation of the hippocampal map in repeated presentations of the same environment over multiple training days. Initially, the correlation is …
Figure 5
Place field directionality depends on statistics of behavior.
(A) Data from Markus et al., 1995, showing that place field remapping depends on the animal’s direction more when the animal is running in a stereotyped path than when the animal is running in …
Figure 6
Response to cue rotation depends on experimental protocol.
(A) Data from Rotenberg and Muller, 1997. The black curve represents the location of the cue card. The heat map represents the firing rate of a given place cell. On rotation of the cue card by 180°, …
Figure 7
Morph experiments.
(A) Different experimental protocols give different results for the morph experiment. The results in the fourth column show the similarities in population representation of the intermediate morph …
Figure 8
Relationship between Rate Remapping and Partial Remapping.
(A) The Beta distribution is used to illustrate the distribution in remapping responses over the place field population. Examples of the Beta distribution for parameter values (red) and …
Figure 9
Animal-to-animal variability may be the result of animal-specific parameter settings.
(A) Simulations from Figure 3C with different values of . Larger values of alpha lead to a greater tendency to infer a larger number of hidden states, and therefore a faster transition from …
Figure 10
Cue variability should affect remapping behavior.
(A) Two training protocols (cyan and magenta) give (B) qualitatively different hidden state inferences when presented with the same novel observation (red dot in A). The cyan training is drawn from …
