Christopher J. Cueva

Compositional Recombination Relies on a Distributed Cortico-Cerebellar Network

The paper uses fMRI and recurrent neural networks to show that compositional cognition involves specialized cortical regions for components and a distributed cortico-cerebellar network for recombination.

Overcoming sensory-memory interference in working memory circuits

This study identifies mechanisms by which neural networks overcome sensory-memory interference using input gating and tuning inversion, supported by neural data from macaques.

Emergent neural dynamics and geometry for generalization in a transitive inference task

This paper develops and tests neural networks that perform transitive inference, showing they generalize and exhibit behaviors similar to humans, while also revealing inconsistencies between human behavior and certain network solutions.

A morphing map model for place field organization in large environments

This paper introduces a morphing map algorithm to model place field organization in bats, showing shared structure between neural and model-generated maps using hyperbolic geometry.

Natural constraints explain working memory capacity limitations in sensory-cognitive models

This study shows that pre-training sensory regions of neural networks with natural images can explain working memory capacity limitations without parameter fitting or hyperparameter tuning.

Low dimensional dynamics for working memory and time encoding

The paper analyzes neural activity in non-human primates during time delays to understand working memory and time encoding mechanisms, finding that time can be decoded with high precision and neural trajectories are low dimensional.