Robert Kim

The paper identifies how medial frontal cortex integrates semantic content and sensory evidence for decision-making using human neurosurgical data and a vision-language model.

The paper investigates VAP family tether selectivity and dynamics at membrane contact sites, revealing distinct MCS footprints and compensatory mechanisms in cellular homeostasis.

The paper shows that verbal task instructions rapidly reconfigure visual representations in ventral temporal cortex via medial frontal cortex modulation.

The paper shows that human working memory is supported by transient, burst-like neural activity rather than sustained firing, with distinct roles for cell types.

This study shows that the human hippocampus can process oddball tones and language stimuli, including semantic and grammatical features, even under general anesthesia.

This study shows that adding random noise to RNNs improves training and enables stable working memory by increasing synaptic decay time constants in inhibitory units.

The paper trains biologically inspired RNNs on cognitive tasks to study how bottom-up and top-down information is integrated, finding that low-level sensory neurons use disinhibitory circuits to combine these signals dynamically.

This paper trains a spiking RNN model on cognitive tasks to uncover task-specific synaptic and membrane parameters, revealing fast membrane properties and slow synaptic dynamics are critical for stimulus encoding and working memory maintenance.

This study examines whether 2D:4D differs between youth with congenital adrenal hyperplasia and controls, finding no significant difference.

The paper shows that strong inhibitory signaling in spiking neural networks supports stable temporal dynamics and working memory by analyzing a model and primate prefrontal cortex data.

This study compares brain volumes in youth with congenital adrenal hyperplasia to healthy controls, finding reduced volumes in specific prefrontal cortex, amygdala, and hippocampal regions.

This study uses spiking recurrent neural networks to model neuronal timescales in working memory and shows they replicate experimental findings.

The paper presents a framework to construct biologically realistic spiking RNNs by transferring dynamics from continuous rate models, achieving similar performance with minimal additional parameters.