Neuronal selectivity and geometric alignment in the human hippocampus support abstract generalization
Abstract representations allow the brain to extract shared structure across different experiences and generalize knowledge beyond individual situations. Although previous studies have shown that representational geometry plays a critical role in supporting abstraction, it remains unclear how the composition of neuronal populations gives rise to such generalizable representations. Here, we…
The human hippocampus plays a crucial role in abstract generalization by organizing neuronal populations in a geometrically aligned manner. A study investigated how neuronal selectivity drives the emergence of abstract representations by combining computational models with analyses of human hippocampal single-neuron recordings. The researchers manipulated artificial neural populations to examine whether increasing task-related information alone could improve cross-context generalization.
However, they found that while increasing stimulus- and response-selective neurons enhanced encoding strength, it did not enhance generalization across contexts. On the other hand, introducing category-selective neurons boosted cross-context generalization, indicating that the type of information represented by a population is crucial for abstraction.
When analyzing human hippocampal neurons, they discovered that category-like and identity-like neurons both increased stimulus encoding, but category-like neurons produced significantly stronger improvements in cross-context generalization. Further analysis revealed that category-like neurons influenced abstraction by reshaping population geometry.
The most significant geometric property associated with generalization was category-axis alignment across contexts, rather than the strength of category-related separation. Mediation analysis suggested that category-like neurons primarily contribute to abstraction by increasing geometric alignment across contexts. The findings reveal a population-level mechanism that links neuronal selectivity to abstract computation, suggesting that flexible generalization relies not only on increasing neural information but also on organizing information into geometries that preserve task-relevant relationships across changing conditions.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
