Attractor Dynamics in the Hippocampus
Published:
Attractor Dynamics in the Hippocampus
Why do certain partial cues, like familiar smells or locations, trigger such strong, vivid, and detailed memories? More abstractly, why can a minor cue generate such a robust and complete mental representation? The hippocampus’s critical role in memory formation and retrieval is well-documented, but the underlying mechanisms remain a subject of inquiry. Two key processes—pattern separation and pattern completion—are facilitated by different areas within the hippocampus. The dentate gyrus is responsible for pattern separation, creating distinct representations during learning to support spatial pattern differentiation. In contrast, the CA3 region is involved in pattern completion, functioning as an attractor or autoassociative network that allows for coherent representations to emerge from minimal cues, such as a familiar scent(Rolls, 2007)
This paper delves into the CA3 structure’s role in episodic memory formation and recall. The hypothesis that CA3 acts as an autoassociative neural network is based on the intricate circuitry within the hippocampus. It has been shown that projections from the entorhinal cortex’s layer 2 reach the dentate gyrus’s granule cells, which in turn project to CA3 cells via mossy fibers. These connections, though sparse, could form a potent link to the CA3 pyramidal cells. Notably, the CA3 system is characterized by its recurrent collaterals, which extend to other hippocampal cells, forming a cohesive, bilaterally connected network. This architecture suggests that the CA3 networks are pivotal in storing episodic memories and that their extensive connectivity is key to retrieving complete representations from partial cues. CA3’s ability to integrate inputs from diverse cerebral cortex areas enables complex associations, such as recalling an object based on a place-related cue, underscoring its role in associative memory tasks. Empirical evidence supports this model, showing that rats with intact CA3 systems can maintain spatial representations with varying cue availability, whereas those with CA3 lesions exhibit impairments.
However, the Continuous Attractor Networks model offers an alternative explanation, positing that spatial representation is inherently continuous, as indicated by the Gaussian-like firing patterns of place and spatial view cells. This model suggests that a Continuous Attractor Neural Network can sustain neuron firing to depict any location along a continuous dimension, like spatial position or head direction, through excitatory recurrent connections that mirror neuronal spatial relationships.
The concept of attractor dynamics extends to the cellular level, where rapid global remappings in the hippocampus indicate shifts in spatial cell firing patterns across different contexts. This phenomenon shows that hippocampal cell subsets can form distinct place fields in varied environments, with spatial relations among cells not preserved across these settings, leading to orthogonal mapping. The challenge then is understanding how established spatial representations are reactivated to ensure stable maps across encounters. Attractor networks, evidenced by the morph sequence experiments transitioning between square and circular environments, highlight how hippocampal patterns adhere to the original map until a critical switch point prompts the adoption of an alternate map. Attractor dynamics thus anchor stable representations, facilitating learning and behavior generalization across similar experiences(Bellmund et al., 2018).
references:
Bellmund, J. L. S., Gärdenfors, P., Moser, E. I., & Doeller, C. F. (2018). Navigating cognition: Spatial codes for human thinking. Science, 362(6415), eaat6766. https://doi.org/10.1126/science.aat6766 Rolls, E. T. (2007). An attractor network in the hippocampus: Theory and neurophysiology. Learning & Memory, 14(11), 714–731. https://doi.org/10.1101/lm.631207