Biology subjects

Dumont, N. S.-Y.

Publications and source records attributed to Dumont, N. S.-Y..

2 recordsLinked to original sources

bioRxiv2026

Semantic map learning and externalization in an embodied neural agent: A comparison to human behavioral and neural data

All mammals can build internal cognitive maps from sensory input, supporting spatial learning and planning. While rodent studies have shown the mammalian navigation system solves the SLAM (Simultaneous Localization and Mapping) problem, its role in human spatial memory and overt recall are less understood. To investigate this, we adapted a spiking semantic SLAM algorithm for a "Treasure Hunt" task, where human participants navigate a 3D beach in virtual reality and later point to remembered object locations. Our agent integrates networks for bipedal locomotion, vision, memory, and arm control to enable first-person learning of place-object associations, and externalizing that knowledge by pointing and expressing confidence. Comparing model observables to human data, we replicate key behavioral and neural effects: monotonic scaling of accuracy with confidence, and recall-dependence on local field potential power observed in the left hippocampus. This work offers a mechanistic framework linking embodied navigation, memory, and communication in human spatial cognition.

animal behavior and cognition↗

bioRxiv2026

Neural Representational Geometry of Feature Binding Operations

The brain faces the feature binding problem: how are multiple stimulus features and variables combined into coherent representations that support flexible behavior? A key finding from neuroscience is that some brain regions employ factorized representations, where distinct features are encoded in neural state space in such a way that enables independent readout and robust generalization. Various algebraic operations have been proposed to model multi-variable representations, but despite extensive study of their theoretical properties (e.g., capacity, noise robustness), it remains unclear which operations produce the representational geometries observed in neural recordings. We systematically evaluate six binding operations implemented in recurrent spiking neural networks performing a working memory task. We find that only superposition and binding with slot-filler structure produce factorized geometry with favorable scaling, while the alternatives do not. These results provide a taxonomy linking algebraic binding operations to neural representational signatures, offering guidance for both computational modelers and experimentalists.

neuroscience↗