How Cognitive Maps Relate to Sequence Learning

This piece, by Armin Bazarjani, was published on 11/19/24. The original text, by George et al., was published by Nature Communications on 04/22/24.

This paper by George et al. proposes a probabilistic sequence model that can learn cognitive maps from aliased (ambiguous) sensory observations. Essentially, they propose that the mental representation of space, our “cognitive map,” is a latent property of higher-order sequence learning.

Similarly, they can explain diverse hippocampal phenomena through a single-sequence learning mechanism. This is interesting as it first provides a unifying theory for context-specific neural responses like landmark and splitter cells. 

Another key contribution of this method is that their model can learn spatial maps from aliased observations and demonstrates transitive inference across disjoint experiences. In other words, the model can natively handle erroneous or ambiguous observations and still construct a meaningful representation. 

The key implication from this model and their follow-up work (Raju et al. 2024) is the idea that space is constructed, in some sort of latent space, through a sequence of sensory observations. This is somewhat convincing, and it is interesting to see how this explains a wide range of neural phenomena. My main concern is how such a model makes something as simple as planning intractable. Definitely curious to see how this line of work develops.

References

George, D., Rikhye, R. V., Gothoskar, N., Guntupalli, J. S., Dedieu, A., & Lázaro-Gredilla, M. (2021). Clone-structured graph representations enable flexible learning and vicarious evaluation of cognitive maps. Nature communications, 12(1), 2392.

Raju, R. V., Guntupalli, J. S., Zhou, G., Wendelken, C., Lázaro-Gredilla, M., & George, D. (2024). Space is a latent sequence: A theory of the hippocampus. Science Advances, 10(31), eadm8470.

Want to submit a piece? Or trying to write a piece and struggling? Check out the guides here!

Thank you for reading. Reminder: Byte Sized is open to everyone! Feel free to submit your piece. Please read the guides first though.

All submissions to berkan@usc.edu with the header “Byte Sized Submission” in Word Doc format please. Thank you!