Jonathan C. Kao

This study compares how cortical and striatal regions encode locomotor preparation and performance in mice, finding complementary coding schemes in different brain areas.

This paper develops an AI-enhanced non-invasive brain-machine interface (BMI) that improves control performance using a CNN and Kalman filter, enabling users to control cursors and robotic arms.

The paper identifies human single neuron representations of intertemporal decision-making in specific brain regions and their relation to reward preferences.

This study identifies methodological flaws and confounding factors in research comparing large language models to brain activity, showing results are unreliable due to issues like shuffled splits and activation extraction biases.

This paper introduces a reinforcement learning-based BCI simulator that accurately replicates real-time BCI decoder performance for cursor control.

This study analyzes spontaneous pain dynamics in chronic pain patients using high-temporal-resolution LFP signals and identifies associations between theta/alpha and gamma band events with pain and opioid effects.

The paper shows that brain areas form minimal sufficient representations of task inputs through preferential propagation of relevant information during decision-making.

This study identifies the dorsal premammillary nucleus as a central node in the escape network by showing its role in controlling escape vigor from both innate and conditioned threats in mice and suggesting a similar pathway in humans.

The paper identifies threat-approach and threat-avoidance ensembles in the dPAG of mice and shows they encode distinct behavioral states.

The paper decodes real-time decision variables from motor cortex activity and shows their link to choice behavior and decision dynamics.

This paper uses multi-area recurrent neural networks to model decision-making dynamics in the brain, finding that orthogonal representations of decision variables are preferentially propagated through inter-area connections.

This paper examines considerations for using RNNs to model neural dynamics in motor behavior, showing that RNNs can recapitulate neural features through different mechanisms and are robust to noise.

This paper introduces LFADS, a deep learning method to infer single-trial neural population dynamics from spiking data, demonstrating improved accuracy in predicting behavior and extracting dynamics across datasets.