Hyperbolic geometry in brain boosts memory, AI capacity
The hippocampus may use hyperbolic space to store memories efficiently...
A theoretical framework shows that neural population geometry in the hippocampus is hyperbolic. A proposed construction of hippocampal tuning curves statistically induces hyperbolic geometry. The Modern Hopfield Network update rule is shown to compute the minimum mean-squared-error estimator, linking neural decoding and associative memory. A novel associative memory model defined in hyperbolic space yields significantly larger capacity than leading models. Results suggest animals encode spatial information as a latent hyperbolic cognitive map, improving both memory capacity and decoding accuracy.
- Hippocampal neural activity exhibits hyperbolic geometry, as shown empirically and now theoretically grounded.
- Modern Hopfield Network update rule can compute the MMSE estimator, linking decoding and associative memory.
- Novel hyperbolic associative memory model achieves significantly larger capacity than leading Euclidean models.
Why It Matters
Could inspire AI memory systems based on hyperbolic space, enabling far larger and more efficient storage.